WO2023200182A1 - Method and device for differentially diagnosing various degenerative brain diseases on basis of deep-learning reading technology in magnetic resonance imaging - Google Patents
Method and device for differentially diagnosing various degenerative brain diseases on basis of deep-learning reading technology in magnetic resonance imaging Download PDFInfo
- Publication number
- WO2023200182A1 WO2023200182A1 PCT/KR2023/004666 KR2023004666W WO2023200182A1 WO 2023200182 A1 WO2023200182 A1 WO 2023200182A1 KR 2023004666 W KR2023004666 W KR 2023004666W WO 2023200182 A1 WO2023200182 A1 WO 2023200182A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- disease
- diagnostic
- degenerative brain
- information
- alzheimer
- Prior art date
Links
- 230000003412 degenerative effect Effects 0.000 title claims abstract description 155
- 208000014644 Brain disease Diseases 0.000 title claims abstract description 153
- 238000000034 method Methods 0.000 title claims abstract description 89
- 238000002595 magnetic resonance imaging Methods 0.000 title claims abstract description 14
- 238000013135 deep learning Methods 0.000 title abstract description 13
- 238000005516 engineering process Methods 0.000 title abstract description 11
- 208000024827 Alzheimer disease Diseases 0.000 claims abstract description 131
- 208000018737 Parkinson disease Diseases 0.000 claims abstract description 74
- 201000002832 Lewy body dementia Diseases 0.000 claims abstract description 73
- 208000009829 Lewy Body Disease Diseases 0.000 claims abstract description 65
- 238000003745 diagnosis Methods 0.000 claims abstract description 61
- 210000004556 brain Anatomy 0.000 claims description 123
- 239000000090 biomarker Substances 0.000 claims description 83
- 238000009825 accumulation Methods 0.000 claims description 65
- 210000003710 cerebral cortex Anatomy 0.000 claims description 50
- 230000008569 process Effects 0.000 claims description 26
- 230000006870 function Effects 0.000 claims description 19
- 238000012545 processing Methods 0.000 claims description 14
- 210000001320 hippocampus Anatomy 0.000 claims description 10
- 210000001769 parahippocampal gyrus Anatomy 0.000 claims description 10
- 230000000007 visual effect Effects 0.000 claims description 10
- 210000001652 frontal lobe Anatomy 0.000 claims description 9
- 238000012800 visualization Methods 0.000 claims description 9
- 201000010099 disease Diseases 0.000 claims description 8
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 7
- 210000003478 temporal lobe Anatomy 0.000 claims description 7
- 238000012937 correction Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000035508 accumulation Effects 0.000 description 62
- 238000013528 artificial neural network Methods 0.000 description 51
- 238000013527 convolutional neural network Methods 0.000 description 27
- 230000001054 cortical effect Effects 0.000 description 14
- 238000010586 diagram Methods 0.000 description 11
- 238000003748 differential diagnosis Methods 0.000 description 11
- 238000011176 pooling Methods 0.000 description 10
- 238000012549 training Methods 0.000 description 9
- 206010067889 Dementia with Lewy bodies Diseases 0.000 description 8
- 230000002490 cerebral effect Effects 0.000 description 8
- 208000015122 neurodegenerative disease Diseases 0.000 description 8
- 238000010801 machine learning Methods 0.000 description 7
- 230000004770 neurodegeneration Effects 0.000 description 7
- 230000001149 cognitive effect Effects 0.000 description 6
- 238000011160 research Methods 0.000 description 6
- 206010012289 Dementia Diseases 0.000 description 5
- 230000004913 activation Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 5
- 230000003920 cognitive function Effects 0.000 description 5
- 238000002591 computed tomography Methods 0.000 description 5
- 239000000284 extract Substances 0.000 description 5
- 210000003625 skull Anatomy 0.000 description 5
- 238000013136 deep learning model Methods 0.000 description 4
- 230000008451 emotion Effects 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 238000007781 pre-processing Methods 0.000 description 4
- 230000000306 recurrent effect Effects 0.000 description 4
- 208000024891 symptom Diseases 0.000 description 4
- 102000013498 tau Proteins Human genes 0.000 description 4
- 108010026424 tau Proteins Proteins 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 208000000044 Amnesia Diseases 0.000 description 3
- 230000007082 Aβ accumulation Effects 0.000 description 3
- 210000000974 brodmann area Anatomy 0.000 description 3
- 238000002059 diagnostic imaging Methods 0.000 description 3
- 238000009593 lumbar puncture Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 210000002682 neurofibrillary tangle Anatomy 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- 230000002123 temporal effect Effects 0.000 description 3
- 210000001519 tissue Anatomy 0.000 description 3
- 208000031091 Amnestic disease Diseases 0.000 description 2
- 208000037259 Amyloid Plaque Diseases 0.000 description 2
- 102000013455 Amyloid beta-Peptides Human genes 0.000 description 2
- 108010090849 Amyloid beta-Peptides Proteins 0.000 description 2
- 208000028698 Cognitive impairment Diseases 0.000 description 2
- 208000012902 Nervous system disease Diseases 0.000 description 2
- 208000025966 Neurological disease Diseases 0.000 description 2
- 150000001413 amino acids Chemical class 0.000 description 2
- 230000006986 amnesia Effects 0.000 description 2
- 210000001175 cerebrospinal fluid Anatomy 0.000 description 2
- 208000010877 cognitive disease Diseases 0.000 description 2
- 238000010219 correlation analysis Methods 0.000 description 2
- 230000006735 deficit Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- -1 mRNA Chemical class 0.000 description 2
- 238000002610 neuroimaging Methods 0.000 description 2
- 210000000056 organ Anatomy 0.000 description 2
- 230000008520 organization Effects 0.000 description 2
- 238000002600 positron emission tomography Methods 0.000 description 2
- 230000002250 progressing effect Effects 0.000 description 2
- 102000004169 proteins and genes Human genes 0.000 description 2
- 108090000623 proteins and genes Proteins 0.000 description 2
- 238000000611 regression analysis Methods 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000002604 ultrasonography Methods 0.000 description 2
- 241000283690 Bos taurus Species 0.000 description 1
- 206010006100 Bradykinesia Diseases 0.000 description 1
- 241000282472 Canis lupus familiaris Species 0.000 description 1
- 241000283707 Capra Species 0.000 description 1
- 208000024172 Cardiovascular disease Diseases 0.000 description 1
- 241000282693 Cercopithecidae Species 0.000 description 1
- 206010061818 Disease progression Diseases 0.000 description 1
- 241000283086 Equidae Species 0.000 description 1
- 241000282326 Felis catus Species 0.000 description 1
- 229930186217 Glycolipid Natural products 0.000 description 1
- 102000003886 Glycoproteins Human genes 0.000 description 1
- 108090000288 Glycoproteins Proteins 0.000 description 1
- 241000282412 Homo Species 0.000 description 1
- 208000006083 Hypokinesia Diseases 0.000 description 1
- 241000124008 Mammalia Species 0.000 description 1
- 208000026139 Memory disease Diseases 0.000 description 1
- 241001465754 Metazoa Species 0.000 description 1
- 241000699670 Mus sp. Species 0.000 description 1
- 208000002740 Muscle Rigidity Diseases 0.000 description 1
- 238000012879 PET imaging Methods 0.000 description 1
- 241001494479 Pecora Species 0.000 description 1
- 206010034719 Personality change Diseases 0.000 description 1
- 208000001431 Psychomotor Agitation Diseases 0.000 description 1
- 241000700159 Rattus Species 0.000 description 1
- 206010071390 Resting tremor Diseases 0.000 description 1
- 206010038743 Restlessness Diseases 0.000 description 1
- 206010041243 Social avoidant behaviour Diseases 0.000 description 1
- 206010065954 Stubbornness Diseases 0.000 description 1
- 241000282887 Suidae Species 0.000 description 1
- 206010044565 Tremor Diseases 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 210000000467 autonomic pathway Anatomy 0.000 description 1
- 210000004204 blood vessel Anatomy 0.000 description 1
- 210000004958 brain cell Anatomy 0.000 description 1
- 210000005013 brain tissue Anatomy 0.000 description 1
- 201000011529 cardiovascular cancer Diseases 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 210000004720 cerebrum Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001684 chronic effect Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000034994 death Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000007850 degeneration Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 150000002016 disaccharides Chemical class 0.000 description 1
- 230000005750 disease progression Effects 0.000 description 1
- 210000005064 dopaminergic neuron Anatomy 0.000 description 1
- 210000001353 entorhinal cortex Anatomy 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 210000004326 gyrus cinguli Anatomy 0.000 description 1
- 230000005032 impulse control Effects 0.000 description 1
- 230000008449 language Effects 0.000 description 1
- 150000002632 lipids Chemical class 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000006984 memory degeneration Effects 0.000 description 1
- 208000023060 memory loss Diseases 0.000 description 1
- 108020004999 messenger RNA Proteins 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 150000002772 monosaccharides Chemical class 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 210000000653 nervous system Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000016273 neuron death Effects 0.000 description 1
- 102000039446 nucleic acids Human genes 0.000 description 1
- 108020004707 nucleic acids Proteins 0.000 description 1
- 150000007523 nucleic acids Chemical class 0.000 description 1
- 229920001542 oligosaccharide Polymers 0.000 description 1
- 150000002482 oligosaccharides Chemical class 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 230000001144 postural effect Effects 0.000 description 1
- 210000000976 primary motor cortex Anatomy 0.000 description 1
- 102000004196 processed proteins & peptides Human genes 0.000 description 1
- 108090000765 processed proteins & peptides Proteins 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 230000037152 sensory function Effects 0.000 description 1
- 230000001568 sexual effect Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 210000004092 somatosensory cortex Anatomy 0.000 description 1
- 210000000278 spinal cord Anatomy 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 235000000346 sugar Nutrition 0.000 description 1
- 150000008163 sugars Chemical class 0.000 description 1
- 210000000225 synapse Anatomy 0.000 description 1
- 230000007470 synaptic degeneration Effects 0.000 description 1
- 210000000857 visual cortex Anatomy 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
- A61B5/004—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part
- A61B5/0042—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part for the brain
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves
- A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/40—Detecting, measuring or recording for evaluating the nervous system
- A61B5/4076—Diagnosing or monitoring particular conditions of the nervous system
- A61B5/4088—Diagnosing of monitoring cognitive diseases, e.g. Alzheimer, prion diseases or dementia
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
Definitions
- This invention was made under the support of the Ministry of Science and ICT under the project number 1711151124 and project number CAP-18-01-KIST.
- the research management agency for the project was the National Research Council of Science and Technology, and the research project name was "Creative Convergence Research.” Project", the research project name is "Advancement of brain plasticity evaluation technology based on brain function-structural imaging", the host organization is Korea University Industry-Academic Cooperation Foundation, and the research period is 2021.09.17. ⁇ 2022.09.16.
- the present invention relates to a method and device for differential diagnosis of various degenerative brain diseases based on deep learning interpretation technology of magnetic resonance images. More specifically, the present invention relates to a method and device for differential diagnosis of cognitive normal (CN), cognitive normal (CN) based on magnetic resonance images using deep learning technology. It relates to a method and device for differential diagnosis of Alzheimer's Disease (AD), Parkinson's Disease (PD), and Dementia with Lewy bodies (DLB).
- AD Alzheimer's Disease
- PD Parkinson's Disease
- DLB Dementia with Lewy bodies
- Alzheimer's disease is a chronic neurodegenerative disease whose symptoms include memory loss, language and cognitive impairment.
- Alzheimer's disease is neuropathologically characterized by the presence of plaques in brain cells, nervous tissue, and blood vessels, neurofibrillary tangles (NFTs), the presence of amyloid peptides that form amyloid plaques, the presence of tau protein, and synapses. Characterized by damage, etc.
- NFTs neurofibrillary tangles
- the cause of Alzheimer's disease is not fully known, and no cure currently exists.
- Alzheimer's disease is a common form of degenerative brain disease and, along with cardiovascular disease and cancer, is a leading cause of death. As human life expectancy increases, the frequency of Alzheimer's disease is also expected to increase.
- a medical examination is performed, or diagnosis is generally performed through positron emission tomography (PET) or cerebrospinal fluid (CSF) lumbar puncture.
- PET positron emission tomography
- CSF cerebrospinal fluid
- the lumbar puncture (lumbar puncture) method was used to measure the degree of accumulation of biomarkers related to degenerative brain diseases such as amyloid and tau.
- existing methods have the problem of high costs, and invasive methods cause pain to patients.
- the present inventors have developed cognitive normal (CN), Alzheimer's Disease (AD), Parkinson's Disease (PD), and Dementia with Lewy bodies (DLB) based on deep learning reading technology.
- a method for differential diagnosis was developed, and as a result of differential diagnosis using the method according to the present invention, it was confirmed that the accuracy was similar to the actual diagnosis result, and it was confirmed that the accuracy was significantly excellent.
- the purpose of the present invention is to provide a method of providing information for diagnosing degenerative brain diseases from medical image data.
- Another object of the present invention is to provide a computer program that provides information for diagnosing degenerative brain diseases from medical image data.
- Another object of the present invention is to provide a computing device that provides information for diagnosing degenerative brain diseases from medical image data.
- the present invention relates to a method and device for differential diagnosis of various degenerative brain diseases based on deep learning reading technology of magnetic resonance images.
- the method according to the present invention is used to diagnose cognitive normal (CN), Alzheimer's disease (AD) , Parkinson's Disease (PD) and Dementia with Lewy bodies (DLB) can be classified with high accuracy.
- CN cognitive normal
- AD Alzheimer's disease
- PD Parkinson's Disease
- DLB Dementia with Lewy bodies
- One aspect of the present invention is a method of providing information for diagnosing a degenerative brain disease from medical image data by a computing device, comprising: a processing step of generating input data based on medical data including a brain image; A generation step of generating one or more diagnostic information by inputting it into a pre-trained degenerative brain disease diagnostic model; and a classification step of classifying normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia based on the diagnostic information.
- medical data in this specification may include multidimensional medical image data composed of discrete image elements (eg, pixels in a two-dimensional image).
- medical data may include a visible object or a digital representation of that object (e.g., a file corresponding to the pixel output of a CT, MRI detector, etc.).
- Medical data is a subject collected by computed tomography (CT), magnetic resonance imaging (MRI), fundus imaging, ultrasound, or any other medical imaging system known in the art. ) may include medical images.
- degenerative brain disease refers to abnormalities in motor control ability, cognitive function, perceptual function, sensory function, and autonomic nerve function due to a decrease or loss of function of nerve cells, and refers to "degenerative nerve disease”, “neurological disease”, and "nervous function”. It can be used with the same meaning as “degenerative disease.”
- Alzheimer's dementia refers to a neurodegenerative disorder and includes familial and sporadic AD.
- the term Alzheimer's dementia may be used interchangeably with the term Alzheimer's disease.
- Typical symptoms of Alzheimer's dementia in human subjects include mild to severe cognitive impairment, progressive impairment of memory (ranging from mild amnesia to disorientation and severe amnesia), deficits in visuospatial skills, personality changes, poor impulse control, and judgment. Examples include, but are not limited to, lack of confidence, distrust of others, increased stubbornness, restlessness, poor planning skills, poor decision-making, and social withdrawal.
- Characteristic pathologies within brain tissue include extracellular neuronal amyloid plaques, neurofibrillary tangles, neurofibrillary degeneration, granovascular neurodegeneration, synapse loss, accumulation of modified tau protein, and widespread neuronal cell death.
- Parkinson's disease refers to a degenerative brain disease of the nervous system caused by loss of dopaminergic neurons. Resting tremor, rigidity, bradykinesia (slow movement), and postural instability are characteristic, and clinical symptoms generally begin to appear after the age of 60.
- Lewy body dementia is a disease caused by the accumulation of abnormally aggregated neurofibrillary proteins within nerve cells, and is known to be accompanied by Parkinson's disease symptoms.
- subject herein refers to a person having a neurodegenerative disease including Alzheimer's dementia, Parkinson's disease, and/or Lewy body dementia, or a neurodegenerative disease including Alzheimer's dementia, Parkinson's disease, and/or Lewy body dementia. It may be a suspect object (or subject).
- the subject may be a mammal, for example, the subject may be one or more selected from the group consisting of humans, monkeys, dogs, cats, mice, rats, cattle, horses, pigs, goats and sheep. You can.
- the learned degenerative brain disease diagnosis model may be a deep neural network.
- deep neural network in this specification may refer to a neural network that includes a plurality of hidden layers in addition to an input layer and an output layer.
- the term “deep neural network” may be used interchangeably with the terms “neural network,” “network function,” and “neural network” throughout this specification.
- Using deep neural networks it is possible to identify latent structures in data. In other words, it is possible to identify the potential structure of a photo, text, video, voice, or music (e.g., what object is in the photo, what the content and emotion of the text are, what the content and emotion of the voice are, etc.) .
- the deep neural network includes a convolutional neural network (CNN), a recurrent neural network (RNN), a restricted Boltzmann machine (RBM), and a deep trust network (deep neural network). It may include, but is not limited to, belief network (DBN), Q network, U network, Siamese network, etc.
- CNN convolutional neural network
- RNN recurrent neural network
- RBM restricted Boltzmann machine
- deep trust network deep neural network
- It may include, but is not limited to, belief network (DBN), Q network, U network, Siamese network, etc.
- the diagnostic information may include a first diagnostic score for discriminating between normal and degenerative brain diseases and a second diagnostic score for discriminating between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia.
- the degenerative brain disease diagnostic model may include a first diagnostic model that generates a first diagnostic score and a second diagnostic model that generates a second diagnostic score.
- medical data may include brain magnetic resonance imaging (MRI) data.
- MRI brain magnetic resonance imaging
- brain magnetic resonance imaging data may be position-corrected to a reference position.
- input data may include a first slice image and a second slice image.
- the processing step may include a slicing step of generating input data including a first slice image and a second slice image from medical data.
- the first slice image and/or the second slice image may include images of one or more brain regions.
- brain region in this specification refers to a part of the brain that is divided by specific criteria and has a certain volume.
- brain regions can be divided by arbitrary criteria such as region function, correlation, location, cell structure and composition, etc., taking into account the diagnostic purpose and characteristics of the subject, or divided by existing partitioning methods. You can.
- brain regions depending on location and function, include the hippocampus, parahippocampal gyrus, frontal lobe, insula, insula, and primary somatosensory cortex.
- primary motor cortex inferior temporal gyrus, superior temporal gyrus, inferior frontal gyrus, anterior cingulate cortex, perirhinal It can be divided into perirhinal cortex, entorhinal cortex, etc. Alternatively, the brain area may be divided according to Brodmann Area and divided into Brodmann Area 1 to Brodmann Area 52.
- the first slice image may include an image of the hippocampus region and the parahippocampal gyrus region.
- the second slice image may include images of the frontal lobe area, the insula, and the insula area.
- the first diagnostic score may be calculated through the first slice image.
- the second diagnostic score may be calculated through the second slice image.
- the input data includes a first slice image and a second slice image
- the degenerative brain disease diagnostic model includes a first diagnostic model that generates a first diagnostic score and a second diagnostic score that generates a second diagnostic score. It may include two diagnostic models, and the first diagnostic score may be calculated from the first slice image through the first diagnostic model, and the second diagnostic score may be calculated from the second slice image through the second diagnostic model.
- the processing step may include a pre-processing step of removing the skull portion from the medical data image using a skull-less mask.
- the generating step is to calculate a partial score including normal probability, Alzheimer's dementia probability, Parkinson's disease probability, and Lewy body dementia probability information based on the first diagnosis score and the second diagnosis score. It may include steps.
- the method provides a visual image that displays the area causing the first diagnostic score and/or the area causing the second diagnostic score in the input data separately from other areas. It may include a visualization step to create.
- the degenerative brain disease diagnosis model may be supervised learning based on one or more learning images and a guide label that corresponds to the learning images and includes a degenerative brain disease diagnosis result.
- supervised learning may be performed based on a comparison result between learning information generated for a learning image using a degenerative brain disease diagnosis model and a guide label.
- supervised learning may be performed based on a result calculated by substituting learning information and guide labels into a binary cross-entropy loss function.
- supervised learning may be learning by adjusting two or more learning data through a mix-up process.
- the method may include an estimation step of calculating a predicted value for the accumulation value of a biomarker related to degenerative brain disease for each brain region by inputting medical data into a pre-learned accumulation value estimation model.
- the method may include an estimation step of calculating a predicted value for the accumulation value of Alzheimer's dementia-related biomarkers for each brain region by inputting medical data into a previously learned accumulation value estimation model. .
- Alzheimer's dementia-related biomarker in this specification refers to a substance that can detect or diagnose a tissue, organ, or entity suffering from Alzheimer's disease by distinguishing it from a normal tissue, organ, or entity.
- a biomarker related to Alzheimer's disease may refer to a substance that accumulates in the brain region of a subject suffering from Alzheimer's disease.
- Biomarkers related to degenerative brain diseases include proteins or nucleic acids (e.g. mRNA, etc.), lipids, glycolipids, glycoproteins, or sugars (monosaccharides, disaccharides, oligosaccharides, etc.) that show an increase or decrease compared to normal subjects without degenerative brain diseases. ) and the like may include organic biomolecules.
- Amyloid in this specification may be interchangeably referred to as “Amyloid beta” or “A ⁇ ”.
- Amyloid refers to a family of major chemical components found in the brain and spinal cord of patients with various degenerative neurological diseases, including Alzheimer's disease.
- Amyloid is a fragment of beta-amyloid precursor protein (APP) containing a variable number of amino acids, typically 38-43 amino acids.
- APP beta-amyloid precursor protein
- the amount of amyloid accumulated in each brain region may vary depending on the stage of disease progression. Specifically, amyloid may be present in a local area of the brain in the early stages of Alzheimer's disease, but may be present in one or more brain areas or throughout the brain as the disease progresses.
- the method according to an embodiment of the present invention uses cerebral cortex thickness data for each brain region of the subject, through which it is possible to predict or calculate the accumulation value of amyloid present in each brain region, and to predict or calculate the accumulation value of amyloid present in each brain region, and to predict or calculate the amyloid accumulation value in each brain region. It is easy to diagnose the presence or absence of a neurodegenerative disease and at the same time obtain information about the progress of the disease, or to diagnose the disease through this.
- the Alzheimer's dementia-related biomarker may be amyloid and/or tau.
- the Alzheimer's dementia-related biomarker may be amyloid.
- the method may include a correction step of correcting the diagnostic information based on a predicted value of the accumulated value of a biomarker related to Alzheimer's disease.
- the prediction value may be calculated from first cerebral cortex thickness data and weight information obtained from medical data.
- weight information in this specification refers to a specific value or specific value that numerically expresses the degree of association between the degenerative brain disease-related biomarkers or Alzheimer's dementia-related biomarkers accumulated in each brain region and the cerebral cortex thickness of the brain region. means a set of Accordingly, weight information may include a weight or a set of weight values. At this time, the weight is repeatedly derived from the degree of correlation between the accumulation value data of biomarkers related to degenerative brain disease or Alzheimer's dementia measured in multiple subjects and the cerebral cortex thickness measured in the same subject as the subject for which the data was measured. It can be defined accordingly.
- Weight information can be defined by calculating the degree of correlation between the accumulation data of biomarkers related to degenerative brain diseases and the cerebral cortex thickness data using a specific model or model, for example, by performing correlation analysis or regression analysis. It can be. When the weight is defined according to correlation analysis or regression analysis, the weight may be defined as a correlation coefficient or regression coefficient. Weight information may include a plurality of weight values.
- cortical thickness data in this specification refers to data containing the measured thickness of the cortical portion of the cerebrum among the subject's brain regions.
- the cerebral cortical thickness data may include cortical thickness information of one or more brain regions. Accordingly, the cerebral cortical thickness data may include cerebral cortical thickness information for each brain region, or may include cerebral cortical thickness information for the entire brain. It is known that cerebral cortex thickness differs between patients with neurodegenerative diseases and normal people. In the present invention, cerebral cortex thickness can be measured directly through the subject's actual brain, or indirectly through video images, etc. When measured indirectly, image data from magnetic resonance imaging (MRI) may be measured using software.
- MRI magnetic resonance imaging
- the cerebral cortex thickness may be first cerebral cortex thickness data, which is data for predicting the accumulation value of biomarkers related to Alzheimer's disease accumulated in the subject's brain, or the accumulation value of biomarkers related to degenerative brain disease. It may be second cerebral cortex thickness data for deriving weight information for calculating the predicted value.
- weight information may be defined between one or more second cerebral cortex thickness data and one or more biomarker accumulation data.
- biomarker accumulation data in this specification refers to data containing accumulation values of biomarkers related to degenerative brain diseases accumulated in the brain of a specific subject.
- Biomarker accumulation data, along with second cerebral cortex thickness data, can be used to determine a weight that quantifies the degree of association between the accumulation of degenerative brain disease-related biomarkers and cerebral cortex thickness as a specific value.
- the biomarker accumulation data may be accumulation data of Alzheimer's dementia-related biomarkers.
- weight information may be defined between one or more second cerebral cortex thickness data and one or more Alzheimer's dementia-related biomarker accumulation data.
- the method may include a decision step of determining a weight for each brain region by deriving the degree of correlation between the second cerebral cortex thickness data and the biomarker accumulation data.
- Another aspect of the present invention is a computer program stored in a storage medium, which, when executed on one or more processors, performs the following operations for providing information for diagnosing a degenerative brain disease, the operations being:
- a processing operation that generates input data based on medical data including brain images
- a classification operation to classify normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia through diagnostic information. It is a computer program stored in a storage medium.
- the diagnostic information may include a first diagnostic score for discriminating between normal and degenerative brain diseases, and a second diagnostic score for discriminating between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia.
- the degenerative brain disease diagnostic model may include a first diagnostic model that generates a first diagnostic score and a second diagnostic model that generates a second diagnostic score.
- medical data may include brain magnetic resonance imaging (MRI) data.
- MRI brain magnetic resonance imaging
- the input data includes a first slice image and a second slice image
- the first slice image includes an image of the hippocampus region and the parahippocampal gyrus region
- the second slice image includes an image of the hippocampus region and the parahippocampal gyrus region.
- the slice image may include images of the frontal lobe area, insula, and temporal lobe area.
- the generating operation calculates a partial score including normal probability, Alzheimer's dementia probability, Parkinson's disease probability, and Lewy body dementia probability information based on the first diagnostic score and the second diagnostic score.
- the program generates a visual image that displays the area causing the first diagnostic score or the area causing the second diagnostic score in the input data separately from other areas. It may include visualization operations.
- the program includes an estimation operation of inputting medical data into a previously learned accumulation value estimation model to calculate a predicted value for the accumulation value of Alzheimer's dementia-related biomarkers for each brain region; And it may include a correction operation for correcting the diagnostic information based on the predicted value of the accumulated value of the Alzheimer's dementia-related biomarker.
- Another aspect of the present invention is a computing device for providing information for diagnosing a degenerative brain disease, comprising: a processor including one or more cores; and memory; wherein the processor generates input data based on medical data including a brain image, inputs the input data into a pre-learned degenerative brain disease diagnosis model to generate one or more diagnostic information, and diagnoses. It is a computing device that classifies normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia through information.
- the diagnostic information may include a first diagnostic score for discriminating between normal and degenerative brain diseases and a second diagnostic score for discriminating between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia.
- the processor determines the normal probability, Alzheimer's, and Alzheimer's disease based on the first and second diagnosis scores.
- a partial score including information on the probability of sexual dementia, probability of Parkinson's disease, and probability of Lewy body dementia can be calculated.
- the processor generates a visual image that displays the area causing the first diagnostic score or the area causing the second diagnostic score in the input data by distinguishing it from other areas. You can.
- the processor inputs medical data into a previously learned accumulation value estimation model to calculate a predicted value for the accumulation value of Alzheimer's dementia-related biomarkers for each brain region; And the diagnostic information can be corrected based on the predicted value of the accumulated value of Alzheimer's dementia-related biomarkers.
- the present invention relates to a method and device for differential diagnosis of various degenerative brain diseases based on deep learning interpretation technology of magnetic resonance images.
- the method according to the present invention enables diagnosis of various degenerative brain diseases even using only magnetic resonance images, so it is inexpensive. It can provide the information necessary for diagnosing degenerative brain diseases, and can differentially diagnose normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia with high accuracy.
- FIG. 1 is a block diagram of a computing device that performs an operation to provide information for diagnosing a degenerative brain disease according to an embodiment of the present invention.
- Figure 2 is a schematic diagram showing a deep neural network according to an embodiment of the present invention.
- Figure 3 is a block diagram illustrating a process for providing information for diagnosing a degenerative brain disease according to an embodiment of the present invention.
- Figure 4 is a block diagram illustrating a process for providing information for diagnosing a degenerative brain disease according to another embodiment of the present invention.
- Figure 5 is a diagram showing the results of diagnosing normal and degenerative brain disease patients according to an embodiment of the present invention.
- Figure 6 is another diagram showing the results of diagnosing normal and degenerative brain disease patients according to an embodiment of the present invention.
- Figure 7 is a diagram showing the results of analyzing the correlation between the diagnostic score generated according to an embodiment of the present invention and the cognitive function test scores, CDR-SB and MMSE.
- the diagnostic information provides information for diagnosing degenerative brain diseases, including a first diagnostic score for distinguishing between normal and degenerative brain diseases and a second diagnostic score for distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia. method.
- ...unit and “...module” used in the specification refer to a unit that processes at least one function or operation, which may be implemented as hardware, software, or a combination of hardware and software.
- FIG. 1 is a block diagram of a computing device that performs an operation to provide information for diagnosing a degenerative brain disease according to an embodiment of the present invention.
- a computing device 1000 that performs an operation to provide information for diagnosing a degenerative brain disease according to an embodiment may include a processor 100 and a memory 200.
- the processor 100 generates input data based on medical data including brain images, inputs the input data into a pre-learned degenerative brain disease diagnosis model to generate one or more diagnostic information, and uses the diagnostic information to determine normal, You can perform actions to classify Alzheimer's disease, Parkinson's disease, and Lewy body dementia.
- the processor 100 may perform an operation of generating diagnostic information including one or more diagnostic scores. For example, the processor may perform an operation to generate a first diagnostic score for discriminating between normal and degenerative brain disease using a pre-learned degenerative brain disease diagnostic model. The processor may perform an operation to generate diagnostic information including a second diagnostic score for distinguishing Alzheimer's dementia, Parkinson's disease, and Lewy body dementia using a pre-learned degenerative brain disease diagnostic model.
- the processor may perform an operation to classify whether the image included in the input data corresponds to normal or degenerative brain disease through the first diagnosis score, and may classify whether the image included in the input data corresponds to a normal or degenerative brain disease, and through the second diagnosis score, the processor may classify whether the image included in the input data corresponds to a normal or degenerative brain disease, and the processor may classify whether the image included in the input data corresponds to a normal or degenerative brain disease, and the processor may classify whether the image included in the input data corresponds to a normal or degenerative brain disease through the first diagnosis score, and the processor may classify whether the image included in the input data corresponds to a normal or degenerative brain disease, and the second diagnosis score may be used to classify whether the image included in the input data corresponds to a normal or degenerative brain disease.
- Medical data may include brain imaging.
- medical data may include images of one or more brain regions.
- the medical data may include multidimensional medical image data composed of discrete image elements.
- medical data may be collected by computed tomography (CT), magnetic resonance imaging (MRI), fundus imaging, ultrasound, or any other medical imaging system known in the art. It may be a medical image of a subject.
- CT computed tomography
- MRI magnetic resonance imaging
- fundus imaging ultrasound, or any other medical imaging system known in the art. It may be a medical image of a subject.
- the processor 100 may perform an operation to equalize medical data. For example, when generating input data based on medical data including a brain magnetic resonance image, input data is generated by performing an operation to equalize the brain magnetic resonance image by adjusting the intensity of the brain magnetic resonance image. can do. If brain magnetic resonance images are standardized, it may be advantageous for learning a degenerative brain disease diagnostic model, and diagnostic accuracy through the degenerative brain disease diagnostic model may be improved.
- the processor 100 may perform an operation to correct the position of the image included in the medical data. For example, when generating input data based on medical data including brain magnetic resonance images, an operation of generating input data can be performed by aligning each brain magnetic resonance image with a reference template. .
- the processor 100 may perform a preprocessing operation to remove areas unrelated to diagnosis from the medical data. For example, when generating input data based on medical data including brain images, the processor can perform a preprocessing operation to remove the skull portion from the image using a skull-less mask for the medical data. there is.
- the processor 100 When generating input data based on medical data including a brain image, the processor 100 performs an operation of generating input data including a slice image of a brain region based on medical data including a brain image. You can.
- the processor may perform an operation of generating input data including a first slice image and a second slice image based on medical data including a brain image.
- the first slice image and the second slice image may include images of different brain regions.
- the first slice image may include images of the hippocampus region and parahippocampal gyrus region.
- the second slice image may include images of the frontal lobe area, insula, and temporal lobe area.
- the processor 100 calculates the first diagnostic score through the first slice image and the first diagnostic score through the second slice image. This may be calculating a two-diagnosis score.
- the first slice image may include images of the hippocampus region and parahippocampal gyrus region
- the second slice image may include images of the frontal lobe, insula, and temporal lobe regions.
- the first slice image includes images of the hippocampus and parahippocampal gyrus regions, it is possible to improve the accuracy of classifying normal and degenerative brain diseases, and if the second slice image includes images of the frontal lobe, insula, and temporal lobe regions. In this case, classification accuracy between Alzheimer's disease, Parkinson's disease, and Lewy body dementia can be improved.
- the pre-trained degenerative brain disease diagnostic model may include one or more diagnostic models.
- a degenerative brain disease diagnostic model may include a first diagnostic model that generates a first diagnostic score and/or a second diagnostic model that generates a second diagnostic score.
- the processor may perform an operation of generating diagnostic information through one or more learned diagnostic models.
- the first diagnostic model may be a model trained in advance to generate a first diagnostic score for discriminating between normal and degenerative brain disease.
- the second diagnostic model may be a model trained in advance to generate a second diagnostic score for Alzheimer's disease, Parkinson's disease, and Lewy body dementia.
- the processor 100 may generate diagnostic information including a diagnostic score obtained by scoring the results of reading medical data using a degenerative brain disease diagnostic model within a specific numerical range. Accordingly, the first diagnostic score and the second diagnostic score may have numerical values within a specific range. At this time, the diagnostic score may be scored as close to the largest value or as close to the smallest value within a specific molar range depending on the severity of normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia. For example, the first diagnostic score may have a value between 0 and 1. If the result determined through the degenerative brain disease diagnostic model is close to normal, it has a value close to 0, and if it is close to a degenerative brain disease, it has a value close to 1. It can be created to have a value close to .
- the first diagnostic score may be generated to have a value closer to 1 when the degenerative brain disease is progressing severely, and a value farther from 1 when the degenerative brain disease is mildly progressing.
- the first diagnostic score can be generated to have .
- the second diagnostic score may have a value between 0 and 1. For example, if the result determined through the degenerative brain disease diagnosis model is close to Alzheimer's dementia, the second diagnosis score may have a value close to 0, if it is close to Parkinson's disease, it may have a value close to 0.5, and Louis It can have a value close to 1, which is close to corpuscular dementia.
- the second diagnostic score may represent probability information corresponding to Alzheimer's dementia, Parkinson's disease, or Lewy body dementia, and accordingly, the results determined through the degenerative brain disease diagnostic model are Alzheimer's dementia, Parkinson's disease, and Lewy body dementia.
- Dementia can have a probability outcome value that adds up to 1.
- the operation of generating the first diagnostic score through the first diagnostic model and the operation of calculating the second diagnostic score through the second diagnostic model may be performed sequentially.
- the first diagnostic score is first calculated using the first diagnostic model, and then the input data determined as degenerative brain disease is divided into Alzheimer's dementia, Parkinson's disease, and Lewy body dementia.
- the second diagnosis score can be calculated through the second diagnosis model.
- the processor 100 When the processor 100 generates one or more diagnostic information by inputting input data into a pre-learned degenerative brain disease diagnostic model, the processor 100 calculates a first diagnostic score through the first slice image using the first diagnostic model, Using the second diagnosis model, the second diagnosis score can be calculated through the second slice image.
- the pre-trained degenerative brain disease diagnosis model may be a deep neural network (DNN).
- DNN deep neural network
- a deep neural network may include an input layer and an output layer, or may include a plurality of separate hidden layers other than the input layer and the output layer.
- Deep neural networks include convolutional neural network (CNN), recurrent neural network (RNN), restricted Boltzmann machine (RBM), and deep belief network (DBN).
- CNN convolutional neural network
- RNN recurrent neural network
- RBM restricted Boltzmann machine
- DBN deep belief network
- a deep neural network may be a convolutional neural network.
- a convolutional neural network is a type of multi-layer perceptron and may include a neural network including a convolutional layer.
- a convolutional neural network can use weights in the computational process through a neural network.
- a convolutional neural network may consist of one or more convolutional layers and neural network layers combined with them.
- the convolutional layer can extract features from input data using filters.
- the convolutional layer may include a filter and an activation function that changes the filter into a non-linear value.
- a convolutional neural network can process image data by representing it as a matrix with dimensions, and through this, the convolutional neural network can be used to recognize objects in images.
- image data encoded in red, green, and blue can be represented by R, G, and B colors as a two-dimensional matrix, i.e., the color value of each pixel can be a component of the matrix, where The size of may be the same as the size of the image.
- a convolutional neural network may include a pooling layer, and through this, it is possible to utilize input data in a two-dimensional structure.
- a convolutional neural network may include one or more convolutional layers and subsampling layers.
- a subsampling layer is connected to the output of the convolutional layer to simplify the output of the convolutional layer. For example, when inputting the output of a convolutional layer to a pooling layer with a 2*2 average pooling filter, the image can be compressed by outputting the average value included in each 2*2 patch for each pixel of the image. there is.
- the above-described pooling may be a method of outputting the minimum value of a patch or the maximum value of a patch, and any pooling method may be used.
- a convolutional neural network can extract features from a given image by repeatedly performing subsampling processes such as convolutional process and pooling. At this time, the output from the convolutional layer and/or subsampling layer may be input to a fully connected layer.
- a fully connected layer is a layer in which all neurons in one layer are connected to all neurons in neighboring layers.
- the processor 100 may perform a visualization operation to generate a visual image that displays the area responsible for generating diagnostic information in input data by distinguishing it from other areas. If the diagnosis information includes a first diagnosis score for distinguishing between normal and degenerative brain disease and a second diagnosis score for distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia, the processor may determine the cause of generating the first diagnosis score.
- a visualization operation can be performed to generate a visual image that displays the area or the area causing the second diagnostic score by distinguishing it from other areas. The visualization operation is performed by comparing output values and slopes based on the generated diagnostic information, displaying the areas causing normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia in the input data, and displaying them separately from other areas. You can.
- the visual image can be a CAM (Class Activation Map) or Grad-CAM.
- the processor 100 When the processor 100 generates one or more diagnostic information by inputting input data into a pre-learned degenerative brain disease diagnostic model, the normal probability, Alzheimer's dementia probability, Parkinson's disease, and Lewy body dementia are based on the diagnostic information.
- An operation may be performed to calculate a partial score including probability information. Specifically, if the diagnostic information includes a first diagnostic score for distinguishing between normal and degenerative brain disease and a second diagnostic score for distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia, the processor Based on the second diagnosis score, an operation can be performed to calculate a partial score including probability information of normalcy, probability of Alzheimer's disease, Parkinson's disease, and probability of Lewy body dementia.
- the processor may calculate the probability of normalcy (a%) using the first diagnostic score, and calculate the probability of degenerative brain disease [(1-a)%] by subtracting the normal probability from the total probability. Then, using the second diagnosis score, the probability of Parkinson's disease (c%) and Lewy body dementia [(1-b-c)%] can be calculated by subtracting the probability of Alzheimer's dementia from the probability of Alzheimer's disease (b%) and the overall probability. You can. And, combining these, the probability of normalcy (a%), probability of Alzheimer's dementia [(1-a)*b%], Parkinson's disease [(1-a)*(c%)], probability of Lewy body dementia [(1- a)*(1-b-c)%] can be calculated.
- the processor 100 may input medical data into a pre-learned accumulation value estimation model and perform an estimation operation to calculate a predicted value for the accumulation value of a biomarker related to a degenerative brain disease in each brain region. Specifically, the processor may perform an estimation operation to calculate a predicted value for the accumulated value of Alzheimer's dementia-related biomarkers for each brain region.
- the processor 100 may calculate a predicted value for an accumulated value of an Alzheimer's dementia-related biomarker and then perform an operation to correct diagnostic information based on the predicted value for an accumulated value of an Alzheimer's dementia-related biomarker. Specifically, the processor may calculate a predicted value for the accumulated value of the Alzheimer's dementia-related biomarker and then perform an operation to correct the second diagnosis score based on the predicted value for the accumulated value of the Alzheimer's dementia-related biomarker. If a predicted value for the accumulated value of Alzheimer's dementia-related biomarkers is calculated, the accuracy of distinguishing Alzheimer's dementia, Parkinson's disease, and Lewy body dementia through the second diagnostic score can be further improved.
- the second diagnosis score can be corrected to be close to the score indicating Alzheimer's dementia, and if the predictive value of amyloid accumulation is below the reference value In this case, the second diagnosis score can be adjusted to be closer to the score indicating Parkinson's disease or Lewy body dementia.
- the correction of diagnostic information through the accumulated value of Alzheimer's dementia-related biomarkers additionally reflects the predictive value of Alzheimer's dementia-related biomarkers accumulated in the brain region in the diagnostic information, and therefore, the diagnosis and treatment of Alzheimer's dementia based solely on medical imaging data Compared to the process of classifying Parkinson's disease and Lewy body dementia, it is possible to differentially diagnose Alzheimer's disease, Parkinson's disease, and Lewy body dementia more accurately.
- the predicted value can be calculated from first cerebral cortex thickness data and weight information obtained from medical data.
- the weight information is obtained from the subject's cerebral cortical thickness data for predicting the degree of accumulation of Alzheimer's-related biomarkers, separate cerebral cortical thickness data (hereinafter referred to as "second cerebral cortical thickness data"), and second cerebral cortical thickness data. It may be defined between the subject and the biomarker accumulation data measured in the same subject.
- the second cerebral cortex thickness data may be measured from T1-weighted images of magnetic resonance imaging (MRI), and the biomarker accumulation data may be obtained from positron emission tomography (PET) images. It may be.
- MRI magnetic resonance imaging
- PET positron emission tomography
- Cortical thickness data can be measured from medical data using a software program.
- cerebral cortex thickness data can be measured from magnetic resonance images in medical data through software.
- a software program for example, the FreeSurfer program can be used. Therefore, through the above program, a pial/white surface is created from the magnetic resonance image to extract the cerebral cortex portion, the thickness of the corresponding cerebral cortex portion is obtained for each position of each vertex defined on the surface, and each First cerebral cortex thickness data can be generated by calculating the average thickness of the vertex corresponding to the brain region.
- cerebral cortex thickness data may be generated from medical data by a processor.
- the cerebral cortex thickness data may be obtained from medical data in a separate process, and the processor may simply use this to calculate a predicted value for the accumulated value of biomarkers related to Alzheimer's dementia.
- Weight information may be defined between the thickness of the second cerebral cortex of each brain region and the biomarker accumulation data of each brain region.
- the weight information may be a set of numerical values indicating the degree of correlation between the accumulated value of the Alzheimer's dementia-related biomarker in one brain region and the cerebral cortex thickness data in the brain region that is the same or not the same as the one brain region. Accordingly, when calculating the predicted value of the accumulated value of a biomarker related to Alzheimer's disease in one brain region, a plurality of weights may be used.
- a case of predicting the degree of accumulation of Alzheimer's-related biomarkers accumulated in the inferior temporal gyrus (ITG) using first cerebral cortex thickness data obtained from a subject is explained.
- weight information defined between the second cerebral cortex thickness data in the ITG region and the biomarker accumulation data in the ITG region, and the first cerebral cortex thickness data in the ITG region are used to determine the Alzheimer's-related biomarkers accumulated in the ITG region.
- the degree of accumulation can be predicted.
- Alzheimer's dementia-related biomarkers accumulated in the ITG region can be predicted from cerebral cortex thickness data of a plurality of brain regions and weight information between the ITG region and each brain region included in the plurality of brain regions.
- the weight information may be determined between the second cerebral cortex thickness data of each brain region and the biomarker accumulation data of the ITG region.
- the plurality of brain regions may not include the ITG region in some cases. That is, the weight defined between the cerebral cortex thickness of the ITG region and the cerebral cortex thickness of the ITG region and the biomarker accumulation data of the ITG region may not be used.
- the predicted value for the accumulation value of biomarkers related to Alzheimer's dementia is between the cerebral cortex thickness data of one or more brain regions, the cerebral cortex data of each brain region included in the one or more brain regions, and the biomarker accumulation data of the brain region to be predicted. It can be calculated using the weight information defined in .
- the Alzheimer's-related biomarker accumulation value of the brain region to be predicted is the weight defined between the cerebral cortex thickness data of each of one or more brain regions and the biomarker accumulation data of the brain region to be predicted, and included in one or more brain regions. It can be calculated by adding the product of the cerebral cortex thickness of each brain region.
- the predicted value for the accumulation value of Alzheimer's-related biomarkers can be calculated using Equation 1 below.
- Weight information can be defined through a machine learning model or deep learning model.
- a pre-trained model may be used as the machine learning model or deep learning model, or a model established through a learning process through one or more second cerebral cortex thickness data and one or more biomarker accumulation data may be used.
- the machine learning model may include a decision tree, Bayesian network, or support vector machine (SVM) algorithm.
- SVM support vector machine
- the second cerebral cortex thickness data and biomarker accumulation data measured in the same subject may be multiple pairs.
- secondary cerebral cortex thickness data and biomarker accumulation data measured in the same subject are 10 pairs, 20 pairs, 30 pairs, 50 pairs, 100 pairs, 200 pairs, 500 pairs, 1000 pairs, 1500 pairs, and 3000 pairs. Or it could be more than 10,000 pairs.
- the accuracy of the weight may be improved by deriving the degree of correlation between a plurality of second cerebral cortex thickness data and biomarker accumulation data. Accordingly, when more pairs of second cerebral cortex thickness data and biomarker accumulation data are used to define weights, the accuracy of the defined weight information can be improved, and through this, the accuracy of predicting Alzheimer's-related biomarker accumulation through the weight information. can be improved.
- the degenerative brain disease diagnosis model may be supervised learning based on one or more learning images and a guide label that corresponds to the learning images and includes a degenerative brain disease diagnosis result.
- Supervised learning can be performed based on the comparison results of learning information and guide labels generated for learning images using a degenerative brain disease diagnosis model.
- Supervised learning can be performed based on a result calculated by substituting the learning information and the guide label into a loss function.
- supervised learning can be performed based on a result calculated by substituting the learning information and the guide label into a binary cross-entropy loss function.
- Supervised learning may be learning by adjusting two or more learning data through a mix-up process.
- Supervised learning can be performed based on the results of comparison between guide labels input using label smoothing and learning information generated for learning images using a degenerative brain disease diagnosis model. Specifically, when distinguishing between normal and degenerative brain disease, normal is set to 0, degenerative brain disease is set to 1, and when distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia, the guide labels are set as probability values for Alzheimer's dementia, Parkinson's disease, and Lewy body dementia.
- Alzheimer's dementia is set to A
- Parkinson's disease is set to B
- Lewy body dementia is set to B.
- You can proceed with learning by entering dementia as c (a+b+c 1).
- the processor 100 may be comprised of one or more cores, and may include a central processing unit (CPU), a graphics processing unit (GPU), or a tensor processing unit (TPU) of a computing device. It may include a processor for data analysis and deep learning.
- the processor may read a computer program stored in a memory and perform data processing for machine learning according to an embodiment.
- the processor may perform calculations for learning a neural network.
- the processor performs calculations for neural network learning, such as processing input data for learning in deep learning (DL), extracting features from input data, calculating errors, and updating the weights of the neural network using backpropagation. It can be done.
- At least one of the CPU, GPU, and TPU of the processor 110 may process learning of the network function.
- the memory 200 includes a flash memory type, hard disk type, multimedia card micro type, card type memory (for example, SD or XD memory, etc.), RAM (Random Access Memory), SRAM (Static Random Access Memory), ROM (Read-Only Memory; ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, It may include at least one type of storage medium among magnetic disks and optical disks.
- a computing device that performs an operation to provide information for diagnosing a degenerative brain disease includes a network unit for transmitting and receiving brain images and other data as needed, and storing and managing data used in other computing processes. Additional databases may be included. Accordingly, a computing device that performs an operation to provide information for diagnosing a degenerative brain disease according to an embodiment may be capable of wired or wireless communication with brain imaging devices, including an MRI imaging device and a PET imaging device. Additionally, data such as weight data, cerebral cortex thickness data, and biomarker accumulation data can be stored in a database, and data stored in the database can be managed, such as deletion and modification, by external input authorized by the user.
- Figure 2 is a schematic diagram showing a deep neural network according to an embodiment of the present invention.
- a neural network may represent a model of a machine learning structure designed to extract feature data from input data and provide inference operations using feature data.
- the feature data may represent data about features in which input data is abstracted.
- the hidden layer is shown as including three layers, but the hidden layer may include a varying number of layers.
- a neural network may include one or more layers, and each layer may include one or more nodes.
- a node is an element that constitutes each layer, and each layer may be composed of a node or a set of nodes.
- nodes in layers other than the output layer can be connected to nodes in the next layer through links for transmitting output signals.
- the nodes of each layer may be connected to each other through a link, and the nodes of the connected layers may be in a relationship as an input node and an output node depending on whether signals are transmitted or received.
- the value of the data of the output node may be determined according to the data input to the input node.
- the output of an activation function regarding the weighted inputs of nodes included in the previous layer may be input to each node included in the hidden layer.
- Weighted input reflects the weight of the input of nodes included in the previous layer.
- the weight may be variable and may vary depending on the function and algorithm of the neural network.
- Weights may be referred to as parameters of the neural network, and activation functions may include sigmoid, hyperbolic tangent (tanh), and rectified linear unit (ReLU).
- the initial input node may refer to one or more nodes in the neural network through which data is directly input without going through links in relationships with other nodes.
- a neural network network in the relationship between nodes based on links, it may mean nodes that do not have other input nodes connected by links.
- the final output node may refer to one or more nodes that do not have an output node in their relationship with other nodes among the nodes in the neural network.
- hidden nodes may refer to nodes constituting a neural network other than the first input node and the last output node.
- a deep neural network may refer to a neural network that includes a plurality of hidden layers in addition to an input layer and an output layer. Deep neural networks allow you to identify latent structures in data. In other words, it is possible to identify the potential structure of a photo, text, video, voice, or music (e.g., what object is in the photo, what the content and emotion of the text are, what the content and emotion of the voice are, etc.) .
- Deep neural networks include convolutional neural networks (CNN), recurrent neural networks (RNN), auto encoders, Generative Adversarial Networks (GAN), restricted Boltzmann machines; RBM), deep belief network (DBN), Q network, U network, Siamese network, Generative Adversarial Network (GAN), etc.
- CNN convolutional neural networks
- RNN recurrent neural networks
- GAN Generative Adversarial Networks
- RBM restricted Boltzmann machines
- DBN deep belief network
- Q network U network
- Siamese network Generative Adversarial Network
- GAN Generative Adversarial Network
- a convolutional neural network is a type of deep learning model for processing data with grid patterns such as images, inspired by the organization of the visual cortex of animals.
- a convolutional neural network may generally include a convolutional layer, a pooling layer, and a fully connected layer. Convolutional layers and pooling layers can exist repeatedly within a neural network, and input data can be transformed into output through these layers.
- the convolution layer uses a kernel (or mask) to extract features, and the product of each element between each element of the kernel and the input value is calculated and summed at each position to obtain an output value, which is called a feature map. It is referred to as This procedure can be repeated applying multiple kernels to form an arbitrary number of feature maps.
- convolutional and pooling layers perform feature extraction, while fully connected layers map the extracted features to the final output, such as a classification operation.
- Neural networks such as convolutional neural networks can be trained to minimize output errors. Separately from the forward propagation process that extracts values from the input layer to the output layer, within the neural network, the error between the input learning data and the corresponding output value of the neural network is calculated and the nodes of each layer are connected to reduce this error. Backpropagation occurs to update the weights.
- the learning process in a convolutional neural network can be summarized as the process of finding a kernel that extracts an output value with the fewest errors based on given training data. The kernel is the only parameter that is automatically learned during the training process of the convolutional layer.
- the size of the kernel, the number of kernels, padding, etc. are hyperparameters that must be set before starting the training process, and therefore, depending on the size of the kernel, the number of kernels, and the number of convolutional layers and pooling layers, They can be divided into different convolutional neural network models.
- Neural networks use supervised learning using training data in which each training data is labeled with the correct answer, unsupervised learning using training data in which the correct answer is not labeled, semi-supervised learning, or reinforcement. It can be learned in at least one way: reinforcement learning.
- the error can be calculated by comparing the output through the neural network and the label or training data, and the calculated error is back-propagated in the neural network in the reverse direction (i.e., from the output layer to the input layer), and the neural network is transmitted according to the back-propagation.
- the connection weight of each node in each layer of the network may be updated. The amount of change in the connection weight of each updated node may be determined according to the learning rate.
- Overfitting is a phenomenon in which errors increase even as the number of training increases due to excessive learning on training data in a neural network. Overfitting can cause errors in machine learning algorithms to increase, and various optimization methods can be used to prevent such overfitting. To prevent overfitting, methods such as increasing the learning data, regularization, dropout to disable some of the network nodes during the learning process, and use of a batch normalization layer can be applied. You can.
- Figure 3 is a block diagram illustrating a process for providing information for diagnosing a degenerative brain disease according to an embodiment of the present invention.
- a computing device that provides information for diagnosing a degenerative brain disease generates input data based on medical data including a brain image (S101) and converts the input data into a pre-learned degenerative brain disease.
- One or more diagnostic information can be generated by inputting it into the brain disease diagnosis model (S102), and the diagnostic information can be used to classify normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia (S103).
- the computing device may generate diagnostic information including one or more diagnostic scores. Specifically, the computing device may generate diagnostic information including a first diagnostic score for discriminating between normal and degenerative brain diseases and a second diagnostic score for discriminating between Alzheimer's disease, Parkinson's disease, and Lewy body dementia.
- the computing device may standardize the medical data. For example, when generating input data based on medical data including a brain magnetic resonance image, the input data can be generated by equalizing the brain magnetic resonance image by adjusting the intensity of the brain magnetic resonance image. If brain magnetic resonance images are standardized, it may be advantageous for learning a degenerative brain disease diagnostic model, and diagnostic accuracy through the degenerative brain disease diagnostic model may be improved.
- the computing device may correct the position of the image included in the medical data. For example, when generating input data based on medical data including a brain magnetic resonance image, the input data can be generated by aligning the position of each brain magnetic resonance image with a reference template.
- the computing device may preprocess the medical data to remove areas unrelated to diagnosis.
- the processor When the processor generates input data based on medical data including brain images, it can perform a preprocessing process to remove the skull portion from the image by using a skull-less mask for the medical data. there is.
- the computing device may generate input data including a slice image of a brain region based on the medical data including a brain image.
- the computing device may generate input data including a first slice image and a second slice image based on medical data including a brain image.
- the first slice image and the second slice image may include images of different brain regions.
- the first slice image may include images of the hippocampus region and parahippocampal gyrus region.
- the second slice image may include images of the frontal lobe area, insula, and temporal lobe area.
- the computing device When inputting input data into a pre-learned degenerative brain disease diagnosis model to generate one or more diagnostic information, the computing device calculates a first diagnosis score through the first slice image and performs a second diagnosis through the second slice image. Scores can be calculated.
- the first slice image may include images of the hippocampus region and parahippocampal gyrus region
- the second slice image may include images of the frontal lobe, insula, and temporal lobe regions.
- the pre-trained degenerative brain disease diagnostic model may include one or more diagnostic models.
- a degenerative brain disease diagnostic model may include a first diagnostic model that generates a first diagnostic score and a second diagnostic model that generates a second diagnostic score.
- the computing device may perform an operation of generating diagnostic information through one or more learned diagnostic models.
- the first diagnostic model may be a model trained in advance to generate a first diagnostic score for discriminating between normal and degenerative brain disease.
- the second diagnosis model may be a model trained in advance to generate a second diagnosis score for distinguishing Alzheimer's disease, Parkinson's disease, and Lewy body dementia.
- the computing device may sequentially perform the process of generating a first diagnostic score through a first diagnostic model and calculating a second diagnostic score through a second diagnostic model.
- the first diagnostic score is first calculated using the first diagnostic model, and then the input data determined as degenerative brain disease is divided into Alzheimer's dementia, Parkinson's disease, and Lewy body dementia.
- the second diagnosis score can be calculated through the second diagnosis model.
- the pre-trained degenerative brain disease diagnosis model may be a deep neural network (DNN).
- DNN deep neural network
- a deep neural network may include an input layer and an output layer, or may include a plurality of separate hidden layers other than the input layer and the output layer.
- Deep neural networks include convolutional neural network (CNN), recurrent neural network (RNN), restricted Boltzmann machine (RBM), and deep belief network (DBN).
- CNN convolutional neural network
- RNN recurrent neural network
- RBM restricted Boltzmann machine
- DBN deep belief network
- a deep neural network may be a convolutional neural network.
- the computing device may perform a visualization operation to generate a visual image that displays the area responsible for generating diagnostic information from input data by distinguishing it from other areas. If the diagnostic information includes a first diagnostic score for distinguishing between normal and degenerative brain disease and a second diagnostic score for distinguishing between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia, the computing device determines the cause of generating the first diagnostic score. Visualization can be performed to generate a visual image that displays the area that is affected or the area that causes the second diagnostic score to be distinguished from other areas. The visualization operation is performed by comparing output values and slopes based on the generated diagnostic information, displaying the areas causing normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia in the input data, and displaying them separately from other areas.
- the visual image can be a CAM (Class Activation Map) or Grad-CAM.
- the computing device When the computing device generates one or more diagnostic information by inputting input data into a pre-learned degenerative brain disease diagnostic model, normal probability, Alzheimer's dementia probability, Parkinson's disease, and Lewy body dementia probability information are generated based on the diagnostic information.
- a partial score including can be calculated (S104). Specifically, when the diagnostic information includes a first diagnostic score for distinguishing between normal and degenerative brain disease and a second diagnostic score for distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia, the computing device may store the first diagnostic score and the above. Based on the second diagnosis score, a partial score including information on the probability of normalcy, probability of Alzheimer's disease, Parkinson's disease, and Lewy body dementia can be calculated.
- Figure 4 is a block diagram illustrating a process for providing information for diagnosing a degenerative brain disease according to another embodiment of the present invention.
- a computing device that provides information for diagnosing a degenerative brain disease generates input data based on medical data including a brain image (S201), and inputs the input data.
- One or more diagnostic information is generated by inputting it into a pre-learned degenerative brain disease diagnosis model (S202), and in order to correct normal and diagnostic information through the diagnostic information, medical data is input into a pre-learned accumulation value estimation model to determine each brain region.
- a predicted value for the accumulated value of biomarkers related to Alzheimer's dementia can be calculated (S205), and Alzheimer's dementia, Parkinson's disease, and Lewy body dementia can be classified (S204).
- the weight information may be defined between the thickness of the second cerebral cortex of each brain region and the biomarker accumulation data of each brain region.
- the weight information may be a set of numerical values indicating the degree of correlation between the accumulated value of the Alzheimer's dementia-related biomarker in one brain region and the cerebral cortex thickness data in the brain region that is the same or not the same as the one brain region. Accordingly, when calculating the predicted value of the accumulated value of a biomarker related to Alzheimer's disease in one brain region, a plurality of weights may be used.
- the predicted value for the accumulation value of biomarkers related to Alzheimer's dementia is between the cerebral cortex thickness data of one or more brain regions, the cerebral cortex data of each brain region included in the one or more brain regions, and the biomarker accumulation data of the brain region to be predicted. It can be calculated using the weight information defined in .
- the Alzheimer's-related biomarker accumulation value of the brain region to be predicted is the weight defined between the cerebral cortex thickness data of each of one or more brain regions and the biomarker accumulation data of the brain region to be predicted, and included in one or more brain regions. It can be calculated by adding the product of the cerebral cortex thickness of each brain region.
- the predicted value for the accumulation value of Alzheimer's-related biomarkers can be calculated using Equation 2 below.
- the computing device may correct the diagnostic information through the calculated predicted value of the accumulated value of the Alzheimer's dementia-related biomarker (S203).
- the second diagnostic score for distinguishing between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia can be corrected through the predicted value of the accumulated value of biomarkers related to Alzheimer's dementia. For example, if the predicted value is above the reference value, the computing device may correct the second diagnosis score to be closer to the score indicating Alzheimer's dementia. Alternatively, if the predicted value is below the reference value, the computing device may correct the second diagnosis score to be closer to the score representing Parkinson's disease and Lewy body dementia.
- Example 1 Differential diagnosis of normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia based on deep learning
- N4 Bias Field Correction was performed, and the intensity of each magnetic resonance image was adjusted to make it uniform.
- ANTS Advanced Normalization Tools
- slice images of brain regions were extracted from the magnetic resonance images using ITK-snap software.
- the slice image for distinguishing between normal and degenerative brain diseases is a cross section based on the z-119 axis in the axial direction
- the slice image for classifying Alzheimer's dementia, Parkinson's disease, and Lewy body dementia is a cross section based on the z-119 axis in the coronal direction.
- a cross section based on the axis was used.
- the skull part of the extracted image was removed using a mask without the skull, and the black background part was cut out to use the part of the image with the brain image.
- the size of the image was changed to fit the model's input value, and normalization was performed to make the distribution of the image uniform.
- the model used is Inception ResNet v2. (CNN classification model), the model receives an image of 299x299x3, extracts features of 35x35x256, goes through Inception A, B, C, and finally goes through sigmoid (in the case of binary classification).
- preprocessed images were input using mix-up, which is a method of combining two random images.
- mix-up is a method of combining two random images.
- label smoothing was used and values of 0.1 instead of 0 and 0.9 instead of 1 were entered.
- the classification between normal people and degenerative brain disease was 0.1 for normal people and 0.9 for degenerative brain disease.
- the learning rate was 0.025, learning was conducted using a total of 5 folds, and the loss function was the Binary Cross-Entropy loss function.
- the method according to one embodiment can distinguish between degenerative brain disease and normal brain disease.
- the prediction score for degenerative brain disease diagnosis from the magnetic resonance image was obtained using the ensemble method using the models obtained in Example 1. Afterwards, the correlation between the obtained degenerative brain disease diagnosis prediction score and the cognitive function test score was calculated.
- the ADNI cohort showed a strong positive/negative linear relationship of 0.7612 for CDR-SB and -0.7435 for MMSE, respectively. Therefore, the method according to the present invention not only differentially diagnoses normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia, but also can estimate the degree of progression of the degenerative brain disease when the degenerative brain disease has progressed through the generated diagnostic score. Confirmed.
- the present invention relates to a method and device for differential diagnosis of various degenerative brain diseases based on deep learning interpretation technology of magnetic resonance images. More specifically, the present invention relates to a method and device for differential diagnosis of cognitive normal (CN), cognitive normal (CN) based on magnetic resonance images using deep learning technology. It relates to a method and device for differential diagnosis of Alzheimer's Disease (AD), Parkinson's Disease (PD), and Dementia with Lewy bodies (DLB).
- AD Alzheimer's Disease
- PD Parkinson's Disease
- DLB Dementia with Lewy bodies
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Molecular Biology (AREA)
- Pathology (AREA)
- Biophysics (AREA)
- Radiology & Medical Imaging (AREA)
- Primary Health Care (AREA)
- Epidemiology (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Surgery (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Neurology (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- High Energy & Nuclear Physics (AREA)
- Child & Adolescent Psychology (AREA)
- Developmental Disabilities (AREA)
- Hospice & Palliative Care (AREA)
- Psychiatry (AREA)
- Psychology (AREA)
- Neurosurgery (AREA)
- Physiology (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The present invention relates to a method and a device for differentially diagnosing various degenerative brain diseases on the basis of deep-learning reading technology in magnetic resonance imaging, and the method according to the present invention can diagnose various degenerative brain diseases even if using only magnetic resonance imaging, and thus can provide, at low cost, information required for diagnosing degenerative brain diseases, and can differentially make a highly accurate diagnosis of normal, Alzheimer's disease, and Parkinson's disease, Lewy body dementia.
Description
본 발명은 과학기술정보통신부의 지원 하에서 과제고유번호 1711151124, 과제번호 CAP-18-01-KIST에 의해 이루어진 것으로서, 상기 과제의 연구관리전문기관은 국가과학기술연구회, 연구사업명은 "창의형융합연구사업", 연구과제명은 "뇌기능-구조 영상기반 뇌 가소성 평가기술의 고도화", 주관기관은 고려대학교 산학협력단, 연구기간은 2021.09.17. ~ 2022.09.16.이다.This invention was made under the support of the Ministry of Science and ICT under the project number 1711151124 and project number CAP-18-01-KIST. The research management agency for the project was the National Research Council of Science and Technology, and the research project name was "Creative Convergence Research." Project", the research project name is "Advancement of brain plasticity evaluation technology based on brain function-structural imaging", the host organization is Korea University Industry-Academic Cooperation Foundation, and the research period is 2021.09.17. ~ 2022.09.16.
본 발명은 자기공명 영상의 딥러닝 판독 기술에 기반한 다양한 퇴행성 뇌질환 감별진단 방법 및 장치에 관한 것으로, 더욱 상세하게는, 딥러닝 기술을 이용하여 자기공명 영상 기반으로 정상 (Cognitive Normal; CN), 알츠하이머성 치매 (Alzheimer's Disease; AD), 파킨슨병 (Parkinson's Disease; PD) 및 루이소체 치매 (Dementia with Lewy bodies; DLB)를 감별진단하는 방법 및 장치에 관한 것이다.The present invention relates to a method and device for differential diagnosis of various degenerative brain diseases based on deep learning interpretation technology of magnetic resonance images. More specifically, the present invention relates to a method and device for differential diagnosis of cognitive normal (CN), cognitive normal (CN) based on magnetic resonance images using deep learning technology. It relates to a method and device for differential diagnosis of Alzheimer's Disease (AD), Parkinson's Disease (PD), and Dementia with Lewy bodies (DLB).
알츠하이머병 (Alzheimer's disease)은 기억력 감퇴, 언어 및 인지장애 등을 증상으로 하는 만성 신경퇴행성 질환이다. 알츠하이머병은 신경병리학적으로 뇌세포, 신경조직 및 혈관 내 침착물 (plaque)의 존재, 신경섬유 농축제 (neurofibrillary tangles; NFT), 아밀로이드 플라크를 형성하는 아밀로이드 펩타이드의 존재, 타우 단백질의 존재 및 시냅스 손상 등을 특징으로 한다. 알츠하이머병의 원인은 완전히 알려져 있지 않으며, 그 치유법도 현재까지는 존재하지 않는다. 알츠하이머병은 퇴행성 뇌질환의 통상적인 형태이며, 심혈관계 질환 및 암과 함께 주요한 사망 원인이다. 인간의 평균 수명이 증가함에 따라, 알츠하이머병의 빈도 역시 증가할 것으로 예상된다.Alzheimer's disease is a chronic neurodegenerative disease whose symptoms include memory loss, language and cognitive impairment. Alzheimer's disease is neuropathologically characterized by the presence of plaques in brain cells, nervous tissue, and blood vessels, neurofibrillary tangles (NFTs), the presence of amyloid peptides that form amyloid plaques, the presence of tau protein, and synapses. Characterized by damage, etc. The cause of Alzheimer's disease is not fully known, and no cure currently exists. Alzheimer's disease is a common form of degenerative brain disease and, along with cardiovascular disease and cancer, is a leading cause of death. As human life expectancy increases, the frequency of Alzheimer's disease is also expected to increase.
종래에는 알츠하이머성 치매와 파킨슨병, 루이소체 치매를 진단하기 위해 문진을 시행하거나, 진단을 위해 일반적으로 양전자 방출 단층 촬영 영상 (positron emission tomography; PET) 또는 뇌척수액 (cerebrospinal fluid; CSF)에서 요추천자 (lumbar puncture) 방법을 이용하여 아밀로이드 및 타우와 같은 퇴행성 뇌질환 관련 바이오마커의 축적 정도를 측정하는 방법을 이용하였다. 하지만 기존의 방법들은 높은 비용이 발생하는 문제점이 있었으며, 침습적 방법의 경우에는 환자에게 고통을 수반하는 문제가 있었다. Conventionally, to diagnose Alzheimer's dementia, Parkinson's disease, and Lewy body dementia, a medical examination is performed, or diagnosis is generally performed through positron emission tomography (PET) or cerebrospinal fluid (CSF) lumbar puncture. The lumbar puncture (lumbar puncture) method was used to measure the degree of accumulation of biomarkers related to degenerative brain diseases such as amyloid and tau. However, existing methods have the problem of high costs, and invasive methods cause pain to patients.
이러한 배경에서, 높은 비용을 발생시키지 않으면서도, 자기공명영상을 통해 대상을 정상, 알츠하이머성 치매 및 파킨슨병, 루이소체 치매를 정확하게 감별진단할 수 있는 방법에 대한 개발이 요구되고 있는 실정이다.Against this background, there is a need for the development of a method that can accurately differentially diagnose a subject as normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia through magnetic resonance imaging without incurring high costs.
이에 본 발명자들은 딥러닝 판독 기술을 기반으로 정상 (Cognitive Normal; CN), 알츠하이머성 치매 (Alzheimer's Disease; AD), 파킨슨병(Parkinson's Disease; PD), 루이소체 치매(Dementia with Lewy bodies; DLB)를 감별진단하는 방법을 개발하였으며, 본 발명에 따른 방법으로 감별진단한 결과, 실제 진단결과와 유사한 정확도를 나타내는 것을 확인하여, 그 정확도가 월등히 우수한 것을 확인하였다. Accordingly, the present inventors have developed cognitive normal (CN), Alzheimer's Disease (AD), Parkinson's Disease (PD), and Dementia with Lewy bodies (DLB) based on deep learning reading technology. A method for differential diagnosis was developed, and as a result of differential diagnosis using the method according to the present invention, it was confirmed that the accuracy was similar to the actual diagnosis result, and it was confirmed that the accuracy was significantly excellent.
이에, 본 발명의 목적은 의료 영상데이터에서 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법을 제공하는 것이다.Accordingly, the purpose of the present invention is to provide a method of providing information for diagnosing degenerative brain diseases from medical image data.
본 발명의 다른 목적은 의료 영상데이터에서 퇴행성 뇌질환 진단을 위한 정보를 제공하는 컴퓨터프로그램을 제공하는 것이다. Another object of the present invention is to provide a computer program that provides information for diagnosing degenerative brain diseases from medical image data.
본 발명의 또 다른 목적은 의료 영상데이터에서 퇴행성 뇌질환 진단을 위한 정보를 제공하는 컴퓨팅 장치를 제공하는 것이다.Another object of the present invention is to provide a computing device that provides information for diagnosing degenerative brain diseases from medical image data.
본 발명은 자기공명 영상의 딥러닝 판독 기술을 기반으로 다양한 퇴행성 뇌질환 감별진단 방법 및 장치에 관한 것으로, 본 발명에 따른 방법은 정상 (Cognitive Normal; CN), 알츠하이머성 치매 (Alzheimer's Disease; AD), 파킨슨병(Parkinson's Disease; PD) 및 루이소체 치매(Dementia with Lewy bodies; DLB)를 높은 정확도로 분류할 수 있다. The present invention relates to a method and device for differential diagnosis of various degenerative brain diseases based on deep learning reading technology of magnetic resonance images. The method according to the present invention is used to diagnose cognitive normal (CN), Alzheimer's disease (AD) , Parkinson's Disease (PD) and Dementia with Lewy bodies (DLB) can be classified with high accuracy.
이하 본 발명을 더욱 자세히 설명하고자 한다.Hereinafter, the present invention will be described in more detail.
본 발명의 일 양태는, 컴퓨팅 장치에 의해 의료 영상데이터에서 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법에 있어서, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하는 처리 단계, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성하는 생성 단계; 및 진단 정보를 통해 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 분류하는 분류 단계;를 포함하는, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법이다.One aspect of the present invention is a method of providing information for diagnosing a degenerative brain disease from medical image data by a computing device, comprising: a processing step of generating input data based on medical data including a brain image; A generation step of generating one or more diagnostic information by inputting it into a pre-trained degenerative brain disease diagnostic model; and a classification step of classifying normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia based on the diagnostic information.
본 명세서 상의 용어 "의료데이터"는, 이산적 영상 요소들 (예를 들어, 2차원 영상에 있어서는 픽셀)로 구성된 다차원 의료 영상 데이터를 포함하는 것일 수 있다. 구체적으로, 의료데이터는 눈으로 볼 수 있는 대상 또는 (예컨대, CT, MRI 검출기 등의 픽셀 출력에 대응되는 파일과 같은) 그 대상의 디지털 표현물을 포함할 수 있다. 의료데이터는 전산화 단층 촬영 (computed tomography; CT), 자기 공명 영상(magnetic resonance imaging; MRI), 안저 영상, 초음파 또는 본 발명의 기술분야에서 공지된 임의의 다른 의료 영상 시스템의 의하여 수집된 대상 (subject)의 의료 영상을 포함하는 것일 수 있다.The term “medical data” in this specification may include multidimensional medical image data composed of discrete image elements (eg, pixels in a two-dimensional image). Specifically, medical data may include a visible object or a digital representation of that object (e.g., a file corresponding to the pixel output of a CT, MRI detector, etc.). Medical data is a subject collected by computed tomography (CT), magnetic resonance imaging (MRI), fundus imaging, ultrasound, or any other medical imaging system known in the art. ) may include medical images.
본 명세서에서 상의 용어 "퇴행성 뇌질환"은 신경세포의 기능 감소 또는 소실에 의해 운동조절능력, 인지기능, 지각기능, 감각기능 및 자율신경의 기능 이상을 의미하며, "퇴행성 신경질환", "신경퇴행성 질환"과 동일한 의미로 사용될 수 있다. As used herein, the term "degenerative brain disease" refers to abnormalities in motor control ability, cognitive function, perceptual function, sensory function, and autonomic nerve function due to a decrease or loss of function of nerve cells, and refers to "degenerative nerve disease", "neurological disease", and "nervous function". It can be used with the same meaning as “degenerative disease.”
본 명세서상의 용어 "알츠하이머성 치매"는 신경퇴행성 장애를 나타내며 가족성 및 산발성 AD를 포함한다. 용어 알츠하이머성 치매는 용어 알츠하이머병과 상호 교차되어 사용될 수 있다. 인간 대상체에서 알츠하이머성 치매를 나타내는 전형적인 증상은 경도 내지 중증의 인지 장애 치매, 기억의 진행성 손상(경도 건망증으로부터 방향감각 상실 및 중증의 기억 상실까지), 시공간 기술 부족, 성격 변화, 충동 제어 부족, 판단력 부족, 타인에 대한 불신, 고집 증가, 안절부절 못함, 계획 능력 부족, 의사 결정 부족, 및 사회적 철회를 들 수 있으며, 다만 이에 제한되는 것은 아니다. 뇌 조직 내의 특징적인 병리로는 세포외 신경 아밀로이드 플라크, 신경원섬유 엉킴, 신경원섬유 퇴행, 과립혈관 신경 퇴행, 시냅스 손실, 변형된 타우 단백질의 축적 및 광범위한 신경 세포 사멸을 들 수 있다.The term “Alzheimer's dementia” herein refers to a neurodegenerative disorder and includes familial and sporadic AD. The term Alzheimer's dementia may be used interchangeably with the term Alzheimer's disease. Typical symptoms of Alzheimer's dementia in human subjects include mild to severe cognitive impairment, progressive impairment of memory (ranging from mild amnesia to disorientation and severe amnesia), deficits in visuospatial skills, personality changes, poor impulse control, and judgment. Examples include, but are not limited to, lack of confidence, distrust of others, increased stubbornness, restlessness, poor planning skills, poor decision-making, and social withdrawal. Characteristic pathologies within brain tissue include extracellular neuronal amyloid plaques, neurofibrillary tangles, neurofibrillary degeneration, granovascular neurodegeneration, synapse loss, accumulation of modified tau protein, and widespread neuronal cell death.
본 명세서상의 용어, "파킨슨병"은 도파민 신경세포의 소실로 인해 발생하는 신경계의 퇴행성 뇌 질환을 의미한다. 안정떨림, 경직, 운동완만 (운동느림) 및 자세 불안정성이 특징적으로 나타나며, 일반적으로 60세 이후에 임상 증상이 나타나기 시작한다.The term "Parkinson's disease" as used herein refers to a degenerative brain disease of the nervous system caused by loss of dopaminergic neurons. Resting tremor, rigidity, bradykinesia (slow movement), and postural instability are characteristic, and clinical symptoms generally begin to appear after the age of 60.
본 명세서상의 용어, "루이소체 치매'는 신경세포 내에 생기는 비정상적으로 응집된 신경섬유단백질의 축적으로 인해 발생하는 질환으로, 파킨슨병 증상을 동반하는 것으로 알려져 있다. As used herein, the term "Lewy body dementia" is a disease caused by the accumulation of abnormally aggregated neurofibrillary proteins within nerve cells, and is known to be accompanied by Parkinson's disease symptoms.
본 명세서 상의 용어 "대상"은 알츠하이머성 치매, 파킨슨병 및/또는 루이소체 치매를 포함하는 신경퇴행성 질환을 가지거나, 또는 알츠하이머성 치매, 파킨슨병 및/또는 루이소체 치매를 포함하는 신경퇴행성 질환이 의심되는 객체 (또는 피검체)일 수 있다. 본 발명의 일 구현예에서, 대상은 포유류 일 수 있고, 예를 들어, 대상은 인간, 원숭이, 개, 고양이, 마우스, 래트, 소, 말, 돼지, 염소 및 양으로 이루어지는 그룹에서 선택되는 하나 이상일 수 있다.The term "subject" herein refers to a person having a neurodegenerative disease including Alzheimer's dementia, Parkinson's disease, and/or Lewy body dementia, or a neurodegenerative disease including Alzheimer's dementia, Parkinson's disease, and/or Lewy body dementia. It may be a suspect object (or subject). In one embodiment of the invention, the subject may be a mammal, for example, the subject may be one or more selected from the group consisting of humans, monkeys, dogs, cats, mice, rats, cattle, horses, pigs, goats and sheep. You can.
본 발명의 일 구현예에서, 학습된 퇴행성 뇌질환 진단 모델은, 딥 뉴럴 네트워크인 것일 수 있다. In one embodiment of the present invention, the learned degenerative brain disease diagnosis model may be a deep neural network.
본 명세서 상의 용어 "딥 뉴럴 네트워크"는 입력 레이어와 출력 레이어 외에 복수의 히든 레이어를 포함하는 신경망을 의미할 수 있다. 용어 "딥 뉴럴 네트워크"는 본 명세서에 걸쳐 용어 "신경망", "네트워크 함수", "뉴럴 네트워크 (neural network)"와 상호교차되어 사용될 수 있다. 딥 뉴럴 네트워크를 이용하면 데이터의 잠재적인 구조 (latent structures)를 파악하는 것이 가능하다. 즉, 사진, 글, 비디오, 음성, 음악의 잠재적인 구조 (예를 들어, 어떤 물체가 사진에 있는지, 글의 내용과 감정이 무엇인지, 음성의 내용과 감정이 무엇인지 등)를 파악할 수 있다. 본 발명에 있어서, 딥 뉴럴 네트워크는 컨볼루셔널 뉴럴 네트워크 (convolutional neural network; CNN), 리커런트 뉴럴 네트워크(recurrent neural network; RNN), 제한 볼츠만 머신 (restricted boltzmann machine; RBM), 심층 신뢰 네트워크(deep belief network; DBN), Q 네트워크, U 네트워크, 샴 네트워크 등을 포함할 수 있으나, 이에 제한되는 것은 아니다.The term “deep neural network” in this specification may refer to a neural network that includes a plurality of hidden layers in addition to an input layer and an output layer. The term “deep neural network” may be used interchangeably with the terms “neural network,” “network function,” and “neural network” throughout this specification. Using deep neural networks, it is possible to identify latent structures in data. In other words, it is possible to identify the potential structure of a photo, text, video, voice, or music (e.g., what object is in the photo, what the content and emotion of the text are, what the content and emotion of the voice are, etc.) . In the present invention, the deep neural network includes a convolutional neural network (CNN), a recurrent neural network (RNN), a restricted Boltzmann machine (RBM), and a deep trust network (deep neural network). It may include, but is not limited to, belief network (DBN), Q network, U network, Siamese network, etc.
본 발명의 일 구현예에서, 진단 정보는 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수 및 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별을 위한 제2진단점수를 포함하는 것일 수 있다.In one embodiment of the present invention, the diagnostic information may include a first diagnostic score for discriminating between normal and degenerative brain diseases and a second diagnostic score for discriminating between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia.
본 발명의 일 구현예에서, 퇴행성 뇌질환 진단 모델은, 제1진단점수를 생성하는 제1진단 모델 및 제2진단점수를 생성하는 제2진단 모델을 포함하는 것일 수 있다.In one embodiment of the present invention, the degenerative brain disease diagnostic model may include a first diagnostic model that generates a first diagnostic score and a second diagnostic model that generates a second diagnostic score.
본 발명의 일 구현예에서, 의료데이터는, 뇌 자기공명영상 (magnetic resonance imaging; MRI) 데이터를 포함하는 것일 수 있다.In one embodiment of the present invention, medical data may include brain magnetic resonance imaging (MRI) data.
본 발명의 일 구현예에서, 뇌 자기공명영상 데이터는 기준 위치로 위치가 보정된 것일 수 있다.In one embodiment of the present invention, brain magnetic resonance imaging data may be position-corrected to a reference position.
본 발명의 일 구현예에서, 입력 데이터는 제1슬라이스 이미지 및 제2슬라이스 이미지를 포함하는 것일 수 있다.In one implementation of the present invention, input data may include a first slice image and a second slice image.
본 발명의 일 구현예에서, 처리 단계는 의료데이터에서 제1슬라이스 이미지 및 제2슬라이스 이미지를 포함하는 입력 데이터를 생성하는 슬라이싱 단계를 포함하는 것일 수 있다.In one implementation of the present invention, the processing step may include a slicing step of generating input data including a first slice image and a second slice image from medical data.
본 발명의 일 구현예에서, 제1슬라이스 이미지 및/또는 제2슬라이스 이미지는 하나 이상의 뇌영역의 이미지를 포함하는 것일 수 있다.In one embodiment of the present invention, the first slice image and/or the second slice image may include images of one or more brain regions.
본 명세서 상의 용어 "뇌 영역"은 특정 기준으로 구획되어 일정 부피를 갖는 뇌의 일부분을 의미한다. 본 발명에 있어서 뇌 영역은 진단 목적 및 대상의 특성을 고려하여 영역의 기능, 연관성, 위치, 세포의 구조와 구성 등의 임의의 기준으로 나누어질 수 있거나, 또는 기존에 존재하는 구획방법으로 나누어질 수 있다. 예를 들어, 뇌 영역은 위치 및 기능에 따라 해마 (hippocampus), 해마이랑 (parahippocampal gyrus), 전두엽 (frontal lobe), 뇌섬엽 (insula), 뇌섬엽 (insula), 일차 몸감각피질 (primary somatosensory cortex), 일차 운동피질 (primary motor cortex), 아래쪽 측두이랑 (inferior temporal gyrus), 위쪽 측두이랑 (superior temporal gyrus), 아래 전두이랑 (inferior frontal gyrus), 앞쪽 띠다발피질 (anterior cingulate cortex), 후각주위피질 (perirhinal cortex), 내후각피질 (entorhinal cortex) 등으로 구분될 수 있다. 또는, 뇌영역은 브로드만 영역 (Brodmann Area)에 따라 구획되어 브로드만 영역 1 내지 브로드만 영역 52로 구획될 수 있다.The term “brain region” in this specification refers to a part of the brain that is divided by specific criteria and has a certain volume. In the present invention, brain regions can be divided by arbitrary criteria such as region function, correlation, location, cell structure and composition, etc., taking into account the diagnostic purpose and characteristics of the subject, or divided by existing partitioning methods. You can. For example, brain regions, depending on location and function, include the hippocampus, parahippocampal gyrus, frontal lobe, insula, insula, and primary somatosensory cortex. ), primary motor cortex, inferior temporal gyrus, superior temporal gyrus, inferior frontal gyrus, anterior cingulate cortex, perirhinal It can be divided into perirhinal cortex, entorhinal cortex, etc. Alternatively, the brain area may be divided according to Brodmann Area and divided into Brodmann Area 1 to Brodmann Area 52.
본 발명의 일 구현예에서, 제1슬라이스 이미지는 해마 (hippocampus) 영역 및 해마이랑 (parahippocampal gyrus) 영역의 이미지를 포함하는 것일 수 있다.In one embodiment of the present invention, the first slice image may include an image of the hippocampus region and the parahippocampal gyrus region.
본 발명의 일 구현예에서, 제2슬라이스 이미지는 전두엽 (frontal lobe) 영역, 뇌섬엽 (insula) 및 뇌섬엽 (insula) 영역의 이미지를 포함하는 것일 수 있다.In one embodiment of the present invention, the second slice image may include images of the frontal lobe area, the insula, and the insula area.
본 발명의 일 구현예에서, 제1진단점수는 제1슬라이스 이미지를 통해 산출되는 것일 수 있다.In one embodiment of the present invention, the first diagnostic score may be calculated through the first slice image.
본 발명의 일 구현예에서, 제2진단 점수는 제2슬라이스 이미지를 통해 산출되는 것일 수 있다.In one embodiment of the present invention, the second diagnostic score may be calculated through the second slice image.
본 발명의 일 구현예에서, 입력 데이터는 제1슬라이스 이미지 및 제2슬라이스 이미지를 포함하고, 퇴행성 뇌질환 진단 모델은 제1진단점수를 생성하는 제1진단 모델 및 제2진단점수를 생성하는 제2진단 모델을 포함할 수 있고, 제1진단점수는 제1진단 모델을 통해 제1슬라이스 이미지로부터 산출되고, 제2진단점수는 제2진단 모델을 통해 제2슬라이스 이미지로부터 산출되는 것일 수 있다.In one embodiment of the present invention, the input data includes a first slice image and a second slice image, and the degenerative brain disease diagnostic model includes a first diagnostic model that generates a first diagnostic score and a second diagnostic score that generates a second diagnostic score. It may include two diagnostic models, and the first diagnostic score may be calculated from the first slice image through the first diagnostic model, and the second diagnostic score may be calculated from the second slice image through the second diagnostic model.
본 발명의 일 구현예에서, 처리 단계는 의료데이터를 두개골 (skull)이 없는 마스크를 이용하여 영상에서 두개골 부분을 제거하는 전처리 단계를 포함하는 것일 수 있다.In one embodiment of the present invention, the processing step may include a pre-processing step of removing the skull portion from the medical data image using a skull-less mask.
본 발명의 일 구현예에서, 생성 단계는, 제1진단점수 및 제2진단점수를 기초로 정상 확률, 알츠하이머성 치매 확률, 파킨슨병 확률 및 루이소체 치매확률 정보를 포함하는 부분 점수를 산출하는 산출 단계를 포함하는 것일 수 있다.In one embodiment of the present invention, the generating step is to calculate a partial score including normal probability, Alzheimer's dementia probability, Parkinson's disease probability, and Lewy body dementia probability information based on the first diagnosis score and the second diagnosis score. It may include steps.
본 발명의 일 구현예에서, 방법은, 입력데이터에서 제1진단점수를 생성한 원인이 되는 영역 및/또는 제2진단점수를 생성한 원인이 되는 영역을 다른 영역과 구분하여 표시하는 시각 영상을 생성하는 시각화 단계를 포함하는 것일 수 있다.In one implementation of the present invention, the method provides a visual image that displays the area causing the first diagnostic score and/or the area causing the second diagnostic score in the input data separately from other areas. It may include a visualization step to create.
본 발명의 일 구현예에서, 퇴행성 뇌질환 진단 모델은, 하나 이상의 학습영상 및 상기 학습영상에 대응되며 퇴행성 뇌질환 진단 결과를 포함하는 가이드 라벨에 기초하여 지도학습되는 것일 수 있다. In one embodiment of the present invention, the degenerative brain disease diagnosis model may be supervised learning based on one or more learning images and a guide label that corresponds to the learning images and includes a degenerative brain disease diagnosis result.
본 발명의 일 구현예에서, 지도학습은 퇴행성 뇌질환 진단 모델을 이용하여 학습영상에 대해 생성한 학습 정보와 가이드 라벨의 비교 결과에 기초하여 수행되는 것일 수 있다.In one embodiment of the present invention, supervised learning may be performed based on a comparison result between learning information generated for a learning image using a degenerative brain disease diagnosis model and a guide label.
본 발명의 일 구현예에서, 지도학습은, 학습 정보와 가이드 라벨을 이진 크로스 엔트로피 (Binary Cross-Entropy) 손실함수에 대입하여 계산한 결과값에 기초하여 수행되는 것일 수 있다.In one implementation of the present invention, supervised learning may be performed based on a result calculated by substituting learning information and guide labels into a binary cross-entropy loss function.
본 발명의 일 구현예에서, 지도학습은, 둘 이상의 학습 데이터를 혼합하는 믹스업 (mix-up) 과정을 통해 조정하여 학습하는 것일 수 있다.In one implementation of the present invention, supervised learning may be learning by adjusting two or more learning data through a mix-up process.
본 발명의 일 구현예에서, 방법은, 의료데이터를 미리 학습된 축적값 추정 모델에 입력하여 뇌 영역별 퇴행성 뇌질환 관련 바이오마커의 축적값에 대한 예측값을 산출하는 추정 단계를 포함하는 것일 수 있다.In one embodiment of the present invention, the method may include an estimation step of calculating a predicted value for the accumulation value of a biomarker related to degenerative brain disease for each brain region by inputting medical data into a pre-learned accumulation value estimation model. .
본 발명의 일 구현예에서, 방법은, 의료데이터를 미리 학습된 축적값 추정 모델에 입력하여 뇌 영역별 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 산출하는 추정 단계를 포함하는 것일 수 있다. In one embodiment of the present invention, the method may include an estimation step of calculating a predicted value for the accumulation value of Alzheimer's dementia-related biomarkers for each brain region by inputting medical data into a previously learned accumulation value estimation model. .
본 명세서 상의 용어 "알츠하이머성 치매 관련 바이오마커"는 알츠하이머성 치매 질환이 발병한 조직, 기관 또는 개체를 정상 조직, 기관 또는 개체와 구분하여 검출 또는 진단할 수 있는 물질을 의미한다. 구체적으로, 알츠하이머성 치매 관련 바이오마커는 알츠하이머성 치매 질환이 발명한 대상의 뇌 영역에 축적되는 물질을 의미할 수 있다. 퇴행성 뇌질환 관련 바이오마커는 퇴행성 뇌질환이 발병하지 않은 정상 대상에 비하여 증가 또는 감소 양상을 보이는 단백질 또는 핵산 (예: mRNA 등), 지질, 당지질, 당단백질 또는 당 (단당류, 이당류, 올리고당류 등) 등과 같은 유기 생체 분자들을 포함할 수 있다. The term "Alzheimer's dementia-related biomarker" in this specification refers to a substance that can detect or diagnose a tissue, organ, or entity suffering from Alzheimer's disease by distinguishing it from a normal tissue, organ, or entity. Specifically, a biomarker related to Alzheimer's disease may refer to a substance that accumulates in the brain region of a subject suffering from Alzheimer's disease. Biomarkers related to degenerative brain diseases include proteins or nucleic acids (e.g. mRNA, etc.), lipids, glycolipids, glycoproteins, or sugars (monosaccharides, disaccharides, oligosaccharides, etc.) that show an increase or decrease compared to normal subjects without degenerative brain diseases. ) and the like may include organic biomolecules.
본 명세서 상의 용어 "아밀로이드 (Amyloid)"는 "아밀로이드 베타 (Amyloid beta)", 또는 "Aβ"로 상호교차하여 지칭될 수 있다. 아밀로이드는 알츠하이머병을 비롯한 다양한 퇴행성 신경계질환 환자의 뇌, 척수 등에서 발견되는 주요 화학적 성분 계열을 의미한다. 아밀로이드는 다양한 수의 아미노산, 통상 38-43 아미노산을 포함하는 베타-아밀로이드 전구 단백질 (APP)의 절편이다. 알츠하이머병과 같은 신경퇴행성 질환에서, 아밀로이드는 질병의 진행단계에 따라 뇌의 영역별로 축적되는 양이 다를 수 있다. 구체적으로, 아밀로이드는 알츠하이머병의 진행 초기에는 뇌의 국소적인 영역에 존재할 수 있으나, 질병 진행 (Disease progression)에 따라서 하나 이상의 뇌영역에 존재하거나, 뇌의 전역에 존재할 수 있다. 본 발명의 일 실시예에 따른 방법은 대상의 뇌 영역별 대뇌 피질 두께 데이터를 이용하며, 이를 통해 뇌의 각 영역에 존재하는 아밀로이드의 축적값을 예측 또는 산출할 수 있고, 대상에서 알츠하이머성 치매 등의 신경퇴행성 질환의 유무를 진단하는 동시에 질병의 진행정도에 대한 정보를 수득하거나, 이를 통해 질병을 진단하는 것이 용이하다.The term “Amyloid” in this specification may be interchangeably referred to as “Amyloid beta” or “Aβ”. Amyloid refers to a family of major chemical components found in the brain and spinal cord of patients with various degenerative neurological diseases, including Alzheimer's disease. Amyloid is a fragment of beta-amyloid precursor protein (APP) containing a variable number of amino acids, typically 38-43 amino acids. In neurodegenerative diseases such as Alzheimer's disease, the amount of amyloid accumulated in each brain region may vary depending on the stage of disease progression. Specifically, amyloid may be present in a local area of the brain in the early stages of Alzheimer's disease, but may be present in one or more brain areas or throughout the brain as the disease progresses. The method according to an embodiment of the present invention uses cerebral cortex thickness data for each brain region of the subject, through which it is possible to predict or calculate the accumulation value of amyloid present in each brain region, and to predict or calculate the accumulation value of amyloid present in each brain region, and to predict or calculate the amyloid accumulation value in each brain region. It is easy to diagnose the presence or absence of a neurodegenerative disease and at the same time obtain information about the progress of the disease, or to diagnose the disease through this.
본 발명의 일 구현예에서, 알츠하이머성 치매 관련 바이오마커는 아밀로이드 및/또는 타우인 것일 수 있다.In one embodiment of the present invention, the Alzheimer's dementia-related biomarker may be amyloid and/or tau.
본 발명의 일 구현예에서, 알츠하이머성 치매 관련 바이오마커는 아밀로이드인 것일 수 있다.In one embodiment of the present invention, the Alzheimer's dementia-related biomarker may be amyloid.
본 발명의 일 구현예에서, 방법은, 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 기초로 상기 진단 정보를 보정하는 보정 단계를 포함하는 것일 수 있다.In one embodiment of the present invention, the method may include a correction step of correcting the diagnostic information based on a predicted value of the accumulated value of a biomarker related to Alzheimer's disease.
본 발명의 일 구현예에서, 예측 값은, 의료데이터에서 수득된 제1대뇌피질 두께 데이터 및 가중치 정보로부터 산출되는 것일 수 있다.In one embodiment of the present invention, the prediction value may be calculated from first cerebral cortex thickness data and weight information obtained from medical data.
본 명세서 상의 용어 "가중치 정보"는 각각의 뇌 영역에 축적된 퇴행성 뇌질환 관련 바이오마커 또는 알츠하이머성 치매 관련 바이오마커와 뇌 영역의 대뇌 피질 두께 사이의 연관정도를 수치적으로 표현한 특정값 또는 특정값들의 집합을 의미한다. 따라서, 가중치 정보는 가중치 또는 가중치값들의 집합을 포함할 수 있다. 이때, 가중치는 복수의 대상에서 측정된 퇴행성 뇌질환 관련 바이오마커 또는 알츠하이머성 치매 관련 바이오마커 축적값 데이터와 해당 데이터가 측정된 대상과 동일한 대상에서 측정된 대뇌 피질 두께 사이의 연관 정도를 반복적으로 도출함에 따라 정의될 수 있다. 가중치 정보는, 퇴행성 뇌질환 관련 바이오마커 축적 데이터와 대뇌 피질 두께 데이터 사이의 연관정도를 특정 모형 또는 모델을 이용해 산출함에 따라 정의될 수 있고, 예를 들어, 상관분석 또는 회귀분석이 수행됨에 따라 정의될 수 있다. 상관분석 (correlation analysis) 또는 회귀분석 (regression analysis)에 따라 가중치가 정의되는 경우, 가중치는 상관계수 또는 회귀계수로 정의될 수 있다. 가중치 정보는 복수의 가중치 값들을 포함하는 것일 수 있다.The term "weight information" in this specification refers to a specific value or specific value that numerically expresses the degree of association between the degenerative brain disease-related biomarkers or Alzheimer's dementia-related biomarkers accumulated in each brain region and the cerebral cortex thickness of the brain region. means a set of Accordingly, weight information may include a weight or a set of weight values. At this time, the weight is repeatedly derived from the degree of correlation between the accumulation value data of biomarkers related to degenerative brain disease or Alzheimer's dementia measured in multiple subjects and the cerebral cortex thickness measured in the same subject as the subject for which the data was measured. It can be defined accordingly. Weight information can be defined by calculating the degree of correlation between the accumulation data of biomarkers related to degenerative brain diseases and the cerebral cortex thickness data using a specific model or model, for example, by performing correlation analysis or regression analysis. It can be. When the weight is defined according to correlation analysis or regression analysis, the weight may be defined as a correlation coefficient or regression coefficient. Weight information may include a plurality of weight values.
본 명세서 상의 용어 "대뇌 피질 두께 (cortical thickness) 데이터"는 대상의 뇌 영역 중 대뇌의 피질 부분의 두께가 측정된 값을 포함하는 데이터를 의미한다. 대뇌 피질 두께 데이터는 하나 이상의 뇌 영역의 대뇌 피질 두께 정보를 포함하는 것일 수 있다. 따라서, 대뇌 피질 두께 데이터는 뇌의 영역별 대뇌 피질 두께 정보를 포함하거나, 또는 뇌 전역에 대한 대뇌 피질 두께 정보를 포함하는 것일 수 있다. 대뇌 피질 두께는 신경 퇴행성 질환 환자와 정상인 사이에서 차이가 발생하는 것으로 알려져 있다. 본 발명에 있어서 대뇌 피질 두께는 대상의 뇌 실물을 통해 직접적으로 측정되거나, 또는 영상 이미지 등을 통해 간접적으로 측정될 수 있다. 간접적으로 측정되는 경우, 자기 공명 영상 (magnetic resonance imaging; MRI)의 영상 이미지 데이터를 소프트웨어를 이용하여 측정될 수 있다. 일 실시예에 있어서 대뇌 피질 두께는 대상의 뇌에 축적된 알츠하이머성 치매 관련 바이오마커의 축적값을 예측하기 위한 데이터인 제1대뇌 피질 두께 데이터일 수 있고, 또는 퇴행성 뇌질환 관련 바이오마커의 축적값의 예측값을 산출하기 위한 가중치 정보를 도출하기 위한 제2대뇌 피질 두께데이터일 수 있다.The term “cortical thickness data” in this specification refers to data containing the measured thickness of the cortical portion of the cerebrum among the subject's brain regions. The cerebral cortical thickness data may include cortical thickness information of one or more brain regions. Accordingly, the cerebral cortical thickness data may include cerebral cortical thickness information for each brain region, or may include cerebral cortical thickness information for the entire brain. It is known that cerebral cortex thickness differs between patients with neurodegenerative diseases and normal people. In the present invention, cerebral cortex thickness can be measured directly through the subject's actual brain, or indirectly through video images, etc. When measured indirectly, image data from magnetic resonance imaging (MRI) may be measured using software. In one embodiment, the cerebral cortex thickness may be first cerebral cortex thickness data, which is data for predicting the accumulation value of biomarkers related to Alzheimer's disease accumulated in the subject's brain, or the accumulation value of biomarkers related to degenerative brain disease. It may be second cerebral cortex thickness data for deriving weight information for calculating the predicted value.
본 발명의 일 구현예에서, 가중치 정보는 하나 이상의 제2대뇌 피질 두께 데이터와 하나 이상의 바이오마커 축적 데이터 사이에서 정의되는 것일 수 있다.In one embodiment of the present invention, weight information may be defined between one or more second cerebral cortex thickness data and one or more biomarker accumulation data.
본 명세서 상의 용어 "바이오마커 축적 데이터"는 특정 대상의 뇌에 축적된 퇴행성 뇌질환 관련 바이오마커의 축적값을 포함하는 데이터를 의미한다. 바이오마커 축적 데이터는 제2대뇌 피질 두께 데이터와 함께 퇴행성 뇌질환 관련 바이오마커의 축적 정도와 대뇌 피질 두께 사이의 연관정도를 특정한 값으로 수치화한 가중치를 결정하는데 이용될 수 있다. 본 발명에 있어서, 바이오마커 축적 데이터는 알츠하이머성 치매 관련 바이오마커의 축적 데이터일 수 있다. 따라서, 일 실시예에 있어서, 가중치 정보는 하나 이상의 제2대뇌 피질 두께 데이터와 하나 이상의 알츠하이머성 치매 관련 바이오마커 축적 데이터 사이에서 정의되는 것일 수 있다. The term “biomarker accumulation data” in this specification refers to data containing accumulation values of biomarkers related to degenerative brain diseases accumulated in the brain of a specific subject. Biomarker accumulation data, along with second cerebral cortex thickness data, can be used to determine a weight that quantifies the degree of association between the accumulation of degenerative brain disease-related biomarkers and cerebral cortex thickness as a specific value. In the present invention, the biomarker accumulation data may be accumulation data of Alzheimer's dementia-related biomarkers. Accordingly, in one embodiment, weight information may be defined between one or more second cerebral cortex thickness data and one or more Alzheimer's dementia-related biomarker accumulation data.
본 발명의 일 구현예에서, 방법은, 제2대뇌 피질 두께 데이터와 바이오마커 축적 데이터 사이의 연관정도를 도출하여 뇌의 각 영역별 가중치를 결정하는 결정 단계를 포함하는 것일 수 있다. In one embodiment of the present invention, the method may include a decision step of determining a weight for each brain region by deriving the degree of correlation between the second cerebral cortex thickness data and the biomarker accumulation data.
본 발명의 다른 양태는, 저장 매체에 저장된 컴퓨터 프로그램으로서, 컴퓨터 프로그램은 하나 이상의 프로세서에서 실행되는 경우, 퇴행성 뇌질환 진단을 위한 정보를 제공하기 위한 이하의 동작들을 수행하도록 하며, 동작들은: Another aspect of the present invention is a computer program stored in a storage medium, which, when executed on one or more processors, performs the following operations for providing information for diagnosing a degenerative brain disease, the operations being:
뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하는 처리 동작; A processing operation that generates input data based on medical data including brain images;
입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성하는 생성 동작; A generation operation of generating one or more diagnostic information by inputting input data into a pre-learned degenerative brain disease diagnostic model;
및 진단 정보를 통해 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 분류하는 분류 동작;을 포함하는 것인, 저장 매체에 저장된 컴퓨터 프로그램이다.and a classification operation to classify normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia through diagnostic information. It is a computer program stored in a storage medium.
본 발명의 일 구현예에서, 진단 정보는 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수, 및 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별을 위한 제2진단점수를 포함하는 것일 수 있다.In one embodiment of the present invention, the diagnostic information may include a first diagnostic score for discriminating between normal and degenerative brain diseases, and a second diagnostic score for discriminating between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia.
본 발명의 일 구현예에서, 퇴행성 뇌질환 진단 모델은, 제1진단점수를 생성하는 제1진단 모델 및 제2진단점수를 생성하는 제2진단 모델을 포함하는 것일 수 있다.In one embodiment of the present invention, the degenerative brain disease diagnostic model may include a first diagnostic model that generates a first diagnostic score and a second diagnostic model that generates a second diagnostic score.
본 발명의 일 구현예에서, 의료데이터는, 뇌 자기공명영상 (magnetic resonance imaging; MRI) 데이터를 포함하는 것일 수 있다.In one embodiment of the present invention, medical data may include brain magnetic resonance imaging (MRI) data.
본 발명의 일 구현예에서, 입력 데이터는 제1슬라이스 이미지 및 제2슬라이스 이미지를 포함하고, 제1슬라이스 이미지는 해마 (hippocampus) 영역 및 해마이랑 (parahippocampal gyrus) 영역의 이미지를 포함하고, 제2슬라이스 이미지는 전두엽 (frontal lobe) 영역, 뇌섬엽 (insula) 및 측두엽 (temporal lobe) 영역의 이미지를 포함하는 것일 수 있다.In one implementation of the present invention, the input data includes a first slice image and a second slice image, the first slice image includes an image of the hippocampus region and the parahippocampal gyrus region, and the second slice image includes an image of the hippocampus region and the parahippocampal gyrus region. The slice image may include images of the frontal lobe area, insula, and temporal lobe area.
본 발명의 일 구현예에서, 생성 동작은 제1진단점수 및 상기 제2진단점수를 기초로 정상 확률, 알츠하이머성 치매 확률, 파킨슨병 확률, 루이소체 치매 확률 정보를 포함하는 부분 점수를 산출하는 산출 동작을 포함할 수 있다.In one embodiment of the present invention, the generating operation calculates a partial score including normal probability, Alzheimer's dementia probability, Parkinson's disease probability, and Lewy body dementia probability information based on the first diagnostic score and the second diagnostic score. Can include actions.
본 발명의 일 구현예에서, 프로그램은, 입력데이터에서 제1진단점수를 생성한 원인이 되는 영역 또는 제2진단점수를 생성한 원인이 되는 영역을 다른 영역과 구분하여 표시하는 시각 영상을 생성하는 시각화 동작을 포함하는 것일 수 있다.In one implementation of the present invention, the program generates a visual image that displays the area causing the first diagnostic score or the area causing the second diagnostic score in the input data separately from other areas. It may include visualization operations.
본 발명의 일 구현예에서, 프로그램은, 의료데이터를 미리 학습된 축적값 추정 모델에 입력하여 뇌 영역별 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 산출하는 추정 동작; 및 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 기초로 진단 정보를 보정하는 보정 동작을 포함하는 것일 수 있다.In one embodiment of the present invention, the program includes an estimation operation of inputting medical data into a previously learned accumulation value estimation model to calculate a predicted value for the accumulation value of Alzheimer's dementia-related biomarkers for each brain region; And it may include a correction operation for correcting the diagnostic information based on the predicted value of the accumulated value of the Alzheimer's dementia-related biomarker.
본 발명의 또 다른 양태는, 퇴행성 뇌질환 진단을 위한 정보를 제공하기 위한 컴퓨팅 장치로서, 하나 이상의 코어를 포함하는 프로세서; 및 메모리;를 포함하고, 프로세서는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하고, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성하고, 및 진단 정보를 통해 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 분류하는, 컴퓨팅장치이다.Another aspect of the present invention is a computing device for providing information for diagnosing a degenerative brain disease, comprising: a processor including one or more cores; and memory; wherein the processor generates input data based on medical data including a brain image, inputs the input data into a pre-learned degenerative brain disease diagnosis model to generate one or more diagnostic information, and diagnoses. It is a computing device that classifies normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia through information.
본 발명의 일 구현예에서, 진단 정보는 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수 및 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별을 위한 제2진단점수를 포함하는 것일 수 있다.In one embodiment of the present invention, the diagnostic information may include a first diagnostic score for discriminating between normal and degenerative brain diseases and a second diagnostic score for discriminating between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia.
본 발명의 일 구현예에서, 프로세서는, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성할 때, 제1진단점수 및 제2진단점수를 기초로 정상 확률, 알츠하이머성 치매 확률, 파킨슨병 확률, 루이소체 치매확률 정보를 포함하는 부분 점수를 산출할 수 있다.In one embodiment of the present invention, when inputting input data to a pre-learned degenerative brain disease diagnostic model to generate one or more diagnostic information, the processor determines the normal probability, Alzheimer's, and Alzheimer's disease based on the first and second diagnosis scores. A partial score including information on the probability of sexual dementia, probability of Parkinson's disease, and probability of Lewy body dementia can be calculated.
본 발명의 일 구현예에서, 프로세서는, 입력데이터에서 제1진단점수를 생성한 원인이 되는 영역 또는 제2진단점수를 생성한 원인이 되는 영역을 다른 영역과 구분하여 표시하는 시각 영상을 생성할 수 있다.In one implementation of the present invention, the processor generates a visual image that displays the area causing the first diagnostic score or the area causing the second diagnostic score in the input data by distinguishing it from other areas. You can.
본 발명의 일 구현예에서, 프로세서는, 의료데이터를 미리 학습된 축적값 추정 모델에 입력하여 뇌 영역별 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 산출하고; 및 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 기초로 진단 정보를 보정할 수 있다.In one embodiment of the present invention, the processor inputs medical data into a previously learned accumulation value estimation model to calculate a predicted value for the accumulation value of Alzheimer's dementia-related biomarkers for each brain region; And the diagnostic information can be corrected based on the predicted value of the accumulated value of Alzheimer's dementia-related biomarkers.
본 발명은 자기공명 영상의 딥러닝 판독 기술을 기반으로 다양한 퇴행성 뇌질환 감별진단 방법 및 장치에 관한 것으로, 본 발명에 따른 방법은 자기공명 영상만을 이용하더라도 다양한 퇴행성 뇌질환의 진단이 가능하여 저렴한 비용으로 퇴행성 뇌질환진단에 필요한 정보를 제공할 수 있으며, 높은 정확도로 정상, 알츠하이머성 치매, 및 파킨슨병, 루이소체 치매를 감별진단할 수 있다.The present invention relates to a method and device for differential diagnosis of various degenerative brain diseases based on deep learning interpretation technology of magnetic resonance images. The method according to the present invention enables diagnosis of various degenerative brain diseases even using only magnetic resonance images, so it is inexpensive. It can provide the information necessary for diagnosing degenerative brain diseases, and can differentially diagnose normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia with high accuracy.
도 1은 본 발명의 일 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하기 위한 동작을 수행하는 컴퓨팅 장치의 블록구성도를 도시한 도면이다.FIG. 1 is a block diagram of a computing device that performs an operation to provide information for diagnosing a degenerative brain disease according to an embodiment of the present invention.
도 2는 본 발명의 일 실시예에 따른 딥 뉴럴 네트워크를 나타낸 개략도이다.Figure 2 is a schematic diagram showing a deep neural network according to an embodiment of the present invention.
도 3은 본 발명의 일 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하는 과정을 설명하기 위한 블록구성도를 도시한 도면이다.Figure 3 is a block diagram illustrating a process for providing information for diagnosing a degenerative brain disease according to an embodiment of the present invention.
도 4는 본 발명의 또 다른 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하는 과정을 설명하기 위한 블록구성도를 도시한 도면이다.Figure 4 is a block diagram illustrating a process for providing information for diagnosing a degenerative brain disease according to another embodiment of the present invention.
도 5는 본 발명의 일 실시예에 따라 정상과 퇴행성 뇌질환환자를 진단한 결과를 나타내는 도면이다.Figure 5 is a diagram showing the results of diagnosing normal and degenerative brain disease patients according to an embodiment of the present invention.
도 6은 본 발명의 일 실시예에 따라 정상과 퇴행성 뇌질환환자를 진단한 결과를 나타내는 다른 도면이다.Figure 6 is another diagram showing the results of diagnosing normal and degenerative brain disease patients according to an embodiment of the present invention.
도 7은 본 발명의 일 실시예에 따라 생성한 진단점수와 인지기능 검사 점수인 CDR-SB 및 MMSE와의 상관관계를 분석한 결과를 나타낸 도면이다.Figure 7 is a diagram showing the results of analyzing the correlation between the diagnostic score generated according to an embodiment of the present invention and the cognitive function test scores, CDR-SB and MMSE.
컴퓨팅 장치에 의해 의료 영상데이터에서 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법에 있어서, In a method of providing information for diagnosing a degenerative brain disease from medical image data by a computing device,
뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하는 처리 단계;A processing step of generating input data based on medical data including brain images;
상기 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성하는 생성 단계; 및A generation step of generating one or more diagnostic information by inputting the input data into a pre-learned degenerative brain disease diagnostic model; and
상기 진단 정보를 통해 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 분류하는 분류 단계;A classification step of classifying normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia based on the diagnostic information;
를 포함하고, Including,
상기 진단 정보는 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수 및 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별을 위한 제2진단점수를 포함하는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.The diagnostic information provides information for diagnosing degenerative brain diseases, including a first diagnostic score for distinguishing between normal and degenerative brain diseases and a second diagnostic score for distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia. method.
아래에서는 첨부한 도면을 참고로 하여 본 발명의 실시예에 대하여 본 발명이 속하는 기술분야에서 통상의 지식을 가진 자가 용이하게 실시할 수 있도록 상세히 설명한다. 그러나 본 발명은 여러가지 상이한 형태로 구현될 수 있으며 여기에서 설명하는 실시예에 한정되지 않는다. 그리고 도면에서 본 발명을 명확하게 설명하기 위해서 설명과 관계없는 부분은 생략하였으며, 명세서 전체를 통하여 유사한 부분에 대해서는 유사한 도면 부호를 붙였다.Below, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily implement the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In order to clearly explain the present invention in the drawings, parts that are not related to the description are omitted, and similar parts are given similar reference numerals throughout the specification.
명세서 전체에서, 어떤 부분이 어떤 구성 요소를 "포함" 한다고 할 때, 이는 특별히 반대되는 기재가 없는 한 다른 구성 요소를 제외하는 것이 아니라 다른 구성 요소를 더 포함할 수 있는 것을 의미한다. 용어 "및/또는," "그리고/또는"은 그 관련되어 나열되는 항목들의 모든 조합들 및 어느 하나를 포함한다.Throughout the specification, when a part "includes" a certain component, this means that it may further include other components rather than excluding other components, unless specifically stated to the contrary. The term “and/or,” “and/or” includes any one and all combinations of the associated listed items.
또한, 명세서에 기재된 "… 부", "… 모듈"의 용어는 적어도 하나의 기능이나 동작을 처리하는 단위를 의미하며, 이는 하드웨어나 소프트웨어 또는 하드웨어 및 소프트웨어의 결합으로 구현될 수 있다.In addition, the terms “…unit” and “…module” used in the specification refer to a unit that processes at least one function or operation, which may be implemented as hardware, software, or a combination of hardware and software.
본 명세서 상에서 실시예들과 관련되어 설명된 다양한 예시적 논리적 블록들, 구성들, 모듈들, 회로들, 수단들, 로직들, 및 알고리즘 단계들이 전자 하드웨어, 컴퓨터 소프트웨어, 또는 양쪽 모두의 조합들로 구현될 수 있음을 인식해야 한다. 하드웨어 및 소프트웨어의 상호교환성을 명백하게 예시하기 위해, 다양한 예시적 컴포넌트들, 블록들, 구성들, 수단들, 로직들, 모듈들, 회로들, 및 단계들은 그들의 기능성 측면에서 일반적으로 위에서 설명되었다. 그러한 기능성이 하드웨어로 또는 소프트웨어로서 구현되는지 여부는 전반적인 시스템에 부과된 특정 어플리케이션(application) 및 설계 제한들에 달려 있다. 숙련된 기술자들은 각각의 특정 어플리케이션들을 위해 다양한 방법들로 설명된 기능성을 구현할 수 있다. 다만, 그러한 구현의 결정들이 본 개시내용의 영역을 벗어나게 하는 것으로 해석되어서는안된다.The various illustrative logical blocks, components, modules, circuits, means, logic, and algorithm steps described in connection with embodiments herein may be implemented in electronic hardware, computer software, or combinations of both. We must recognize that it can be implemented. To clearly illustrate the interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented in hardware or software will depend on the specific application and design constraints imposed on the overall system. A skilled technician can implement the described functionality in a variety of ways for each specific application. However, such implementation decisions should not be construed as causing a departure from the scope of the present disclosure.
도 1은 본 발명의 일 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하기 위한 동작을 수행하는 컴퓨팅 장치의 블록구성도를 도시한 도면이다.FIG. 1 is a block diagram of a computing device that performs an operation to provide information for diagnosing a degenerative brain disease according to an embodiment of the present invention.
도 1을 참조하면, 일 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하기 위한 동작을 수행하는 컴퓨팅 장치 (1000)는 프로세서 (100) 및 메모리 (200)를 포함할 수 있다. Referring to FIG. 1, a computing device 1000 that performs an operation to provide information for diagnosing a degenerative brain disease according to an embodiment may include a processor 100 and a memory 200.
프로세서 (100)는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하고, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성하고, 진단 정보를 통해 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 분류하는 동작을 수행할 수 있다.The processor 100 generates input data based on medical data including brain images, inputs the input data into a pre-learned degenerative brain disease diagnosis model to generate one or more diagnostic information, and uses the diagnostic information to determine normal, You can perform actions to classify Alzheimer's disease, Parkinson's disease, and Lewy body dementia.
프로세서 (100)는, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성할 때, 하나 이상의 진단 점수를 포함하는 진단 정보를 생성하는 동작을 수행할 수 있다. 예를 들어, 프로세서는 미리 학습된 퇴행성 뇌질환 진단 모델을 이용하여 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수를 생성하기 위한 동작을 수행할 수 있다. 프로세서는 미리 학습된 퇴행성 뇌질환 진단 모델을 이용하여 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별을 위한 제2진단점수를 포함하는 진단 정보를 생성하기 위한 동작을 수행할 수 있다. 이 경우, 프로세서는 제1진단점수를 통해 입력 데이터에 포함된 영상이 정상 또는 퇴행성 뇌질환에 해당되는지 분류하는 동작을 수행할 수 있고, 제2진단점수를 통해 알츠하이머성 치매, 파킨슨병 및 루이소체 치매에 해당되는지 분류하는 동작을 수행할 수 있다.When generating one or more diagnostic information by inputting input data into a pre-learned degenerative brain disease diagnostic model, the processor 100 may perform an operation of generating diagnostic information including one or more diagnostic scores. For example, the processor may perform an operation to generate a first diagnostic score for discriminating between normal and degenerative brain disease using a pre-learned degenerative brain disease diagnostic model. The processor may perform an operation to generate diagnostic information including a second diagnostic score for distinguishing Alzheimer's dementia, Parkinson's disease, and Lewy body dementia using a pre-learned degenerative brain disease diagnostic model. In this case, the processor may perform an operation to classify whether the image included in the input data corresponds to normal or degenerative brain disease through the first diagnosis score, and may classify whether the image included in the input data corresponds to a normal or degenerative brain disease, and through the second diagnosis score, the processor may classify whether the image included in the input data corresponds to a normal or degenerative brain disease, and the processor may classify whether the image included in the input data corresponds to a normal or degenerative brain disease, and the processor may classify whether the image included in the input data corresponds to a normal or degenerative brain disease through the first diagnosis score, and the processor may classify whether the image included in the input data corresponds to a normal or degenerative brain disease, and the second diagnosis score may be used to classify whether the image included in the input data corresponds to a normal or degenerative brain disease. You can perform actions to classify whether a person has dementia.
의료데이터는 뇌 영상을 포함할 수 있다. 예를 들어, 의료데이터는 뇌영역 중 하나 이상의 뇌영역에 대한 영상을 포함하는 것일 수 있다. 이때, 의료데이터는 이산적 영상 요소들로 구성된 다차원 의료 영상 데이터를 포함하는 것일 수 있다. 예를 들어, 의료 데이터는 전산화 단층 촬영 (computed tomography; CT), 자기 공명 영상 (magnetic resonance imaging; MRI), 안저 영상, 초음파 또는 본 발명의 기술분야에서 공지된 임의의 다른 의료 영상 시스템의 의하여 수집된 대상 (subject)의 의료 영상일 수 있다.Medical data may include brain imaging. For example, medical data may include images of one or more brain regions. At this time, the medical data may include multidimensional medical image data composed of discrete image elements. For example, medical data may be collected by computed tomography (CT), magnetic resonance imaging (MRI), fundus imaging, ultrasound, or any other medical imaging system known in the art. It may be a medical image of a subject.
프로세서 (100)는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성할 때, 의료데이터를 균일화하는 동작을 수행할 수 있다. 예를 들어, 뇌 자기 공명 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하는 경우, 뇌 자기 공명 영상의 강도 (intensity)를 조절하여 뇌 자기 공명영상을 균일화하는 동작을 수행하여 입력 데이터를 생성할 수 있다. 뇌 자기 공명영상이 균일화되는 경우, 퇴행성 뇌질환 진단 모델의 학습에 유리할 수 있으며, 퇴행성 뇌질환 진단 모델을 통한 진단 정확도가 향상될 수 있다.When generating input data based on medical data including brain images, the processor 100 may perform an operation to equalize medical data. For example, when generating input data based on medical data including a brain magnetic resonance image, input data is generated by performing an operation to equalize the brain magnetic resonance image by adjusting the intensity of the brain magnetic resonance image. can do. If brain magnetic resonance images are standardized, it may be advantageous for learning a degenerative brain disease diagnostic model, and diagnostic accuracy through the degenerative brain disease diagnostic model may be improved.
프로세서 (100)는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성할 때, 의료데이터에 포함된 영상의 위치를 보정하는 동작을 수행할 수 있다. 예를 들어, 뇌 자기 공명 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하는 경우, 각각의 뇌 자기 공명영상을 기준 템플레이트 (template)로 위치를 맞추어 입력데이터를 생성하는 동작을 수행할 수 있다.When generating input data based on medical data including a brain image, the processor 100 may perform an operation to correct the position of the image included in the medical data. For example, when generating input data based on medical data including brain magnetic resonance images, an operation of generating input data can be performed by aligning each brain magnetic resonance image with a reference template. .
프로세서 (100)는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성할 때, 의료데이터에서 진단과 관련 없는 영역을 제거하는 전처리 동작을 수행할 수 있다. 예를 들어, 프로세서는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성할 때, 의료데이터를 두개골 (skull)이 없는 마스크를 이용하여 영상에서 두개골 부분을 제거하는 전처리 동작을 수행할 수 있다.When generating input data based on medical data including brain images, the processor 100 may perform a preprocessing operation to remove areas unrelated to diagnosis from the medical data. For example, when generating input data based on medical data including brain images, the processor can perform a preprocessing operation to remove the skull portion from the image using a skull-less mask for the medical data. there is.
프로세서 (100)는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성할 때, 뇌 영상을 포함하는 의료데이터를 바탕으로 뇌 영역의 슬라이스 이미지를 포함하는 입력데이터를 생성하는 동작을 수행할 수 있다. 예를 들어, 프로세서는 뇌 영상을 포함하는 의료데이터를 기초로 제1슬라이스 이미지 및 제2슬라이스 이미지를 포함하는 입력 데이터를 생성하는 동작을 수행할 수 있다. 이때, 제1슬라이스 이미지 및 제2슬라이스 이미지에는 각기 다른 뇌영역의 이미지가 포함될 수 있다. 구체적으로, 제1슬라이스 이미지는 해마 (hippocampus) 영역 및 해마이랑 (parahippocampal gyrus) 영역의 이미지를 포함할 수 있다. 제2슬라이스 이미지는 전두엽 (frontal lobe) 영역, 뇌섬엽 (insula) 및 측두엽 (temporal lobe) 영역의 이미지를 포함할 수 있다.When generating input data based on medical data including a brain image, the processor 100 performs an operation of generating input data including a slice image of a brain region based on medical data including a brain image. You can. For example, the processor may perform an operation of generating input data including a first slice image and a second slice image based on medical data including a brain image. At this time, the first slice image and the second slice image may include images of different brain regions. Specifically, the first slice image may include images of the hippocampus region and parahippocampal gyrus region. The second slice image may include images of the frontal lobe area, insula, and temporal lobe area.
프로세서 (100)는, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성할 때, 제1슬라이스 이미지를 통해 제1진단점수를 산출하고, 제2슬라이스 이미지를 통해 제2진단점수를 산출하는 것일 수 있다. 이때, 전술한 바와 같이, 제1슬라이스 이미지에는 해마 영역 및 해마이랑 영역의 이미지가 포함될 수 있고, 제2슬라이스 이미지에는 전두엽, 뇌섬엽 및 측두엽 영역의 이미지가 포함될 수 있다. 제1슬라이스 이미지에 해마 영역 및 해마이랑 영역의 이미지가 포함되는 경우, 정상과 퇴행성 뇌질환 분류의 정확도가 향상되는 것이 가능하며, 제2슬라이스 이미지에 전두엽, 뇌섬엽 및 측두엽 영역의 이미지가 포함되는 경우 알츠하이머성 치매와 파킨슨병, 루이소체 치매 사이의 분류 정확도가 향상될 수 있다.When the processor 100 generates one or more diagnostic information by inputting input data into a pre-learned degenerative brain disease diagnostic model, the processor 100 calculates the first diagnostic score through the first slice image and the first diagnostic score through the second slice image. This may be calculating a two-diagnosis score. At this time, as described above, the first slice image may include images of the hippocampus region and parahippocampal gyrus region, and the second slice image may include images of the frontal lobe, insula, and temporal lobe regions. If the first slice image includes images of the hippocampus and parahippocampal gyrus regions, it is possible to improve the accuracy of classifying normal and degenerative brain diseases, and if the second slice image includes images of the frontal lobe, insula, and temporal lobe regions. In this case, classification accuracy between Alzheimer's disease, Parkinson's disease, and Lewy body dementia can be improved.
일 실시예에서, 미리 학습된 퇴행성 뇌질환 진단 모델은 하나 이상의 진단 모델을 포함하는 것일 수 있다. 예를 들어, 퇴행성 뇌질환 진단 모델은 제1진단점수를 생성하는 제1진단 모델 및/또는 제2진단점수를 생성하는 제2진단 모델을 포함하는 것일 수 있다. 따라서, 프로세서는 하나 이상의 학습된 진단 모델을 통하여 진단 정보를 생성하는 동작을 수행할 수 있다. 이때, 제1진단모델은 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수를 생성하도록 미리 학습된 모델일 수 있다. 제2진단모델은 알츠하이머성 치매, 파킨슨병 및 루이소체 치매을 위한 제2진단점수를 생성하도록 미리 학습된 모델일 수 있다. In one embodiment, the pre-trained degenerative brain disease diagnostic model may include one or more diagnostic models. For example, a degenerative brain disease diagnostic model may include a first diagnostic model that generates a first diagnostic score and/or a second diagnostic model that generates a second diagnostic score. Accordingly, the processor may perform an operation of generating diagnostic information through one or more learned diagnostic models. At this time, the first diagnostic model may be a model trained in advance to generate a first diagnostic score for discriminating between normal and degenerative brain disease. The second diagnostic model may be a model trained in advance to generate a second diagnostic score for Alzheimer's disease, Parkinson's disease, and Lewy body dementia.
프로세서 (100)는, 퇴행성 뇌질환 진단 모델을 통해 의료 데이터를 판독한 결과를 특정 수치범위내에서 점수화한 진단점수를 포함하는 진단 정보를 생성할 수 있다. 따라서, 제1진단점수 및 제2진단점수는 특정 범위의 수치값을 가질 수 있다. 이때, 진단점수는 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매의 중증도에 따라 특정 구치 범위내에서 가장 큰값에 가깝도록 점수화 되거나, 가장 작은 값에 가깝도록 점수화 될 수 있다. 예를 들어, 제1진단점수는 0 과 1 사이의 값을 가질 수 있고, 퇴행성 뇌질환 진단 모델을 통해 판별된 결과가 정상에 가까운 경우 0에 가까운 값을 가지고, 퇴행성 뇌질환에 가까운 경우에는 1에 가까운 값을 갖도록 생성될 수 있다. 이때, 퇴행성 뇌질환의 중증도에 따라, 퇴행성 뇌질환이 진행이 심한 경우에는 1에 보다 가까운 값을 가지도록 제1진단점수가 생성될 수 있고, 퇴행성 뇌질환이 약하게 진행된 경우에는 1에 보다 먼 값을 가지도록 제1진단점수가 생성될 수 있다. 또한, 제2진단점수도 0과 1 사이의 값을 가질 수 있다. 예를 들어, 퇴행성 뇌질환 진단 모델을 통해 판별된 결과가 알츠하이머성 치매에 가까운 경우 제2진단점수는 0에 가까운 값을 가질 수 있고, 파킨슨병에 가까운 경우 0.5에 가까운 값을 가질 수 있으며, 루이소체 치매에 가까운 경 1에 가까운 값을 가질 수 있다. 또는, 제2진단점수는 알츠하이머성 치매, 파킨슨병 또는 루이소체 치매에 해당하는 확률 정보를 나타내는 것일 수 있으며, 이에 따라 퇴행성 뇌질환 진단 모델을 통해 판별된 결과가 알츠하이머성 치매, 파킨슨병 및 루이소체 치매는 도합 1이 되는 확률 결과 값을 가질 수 있다.The processor 100 may generate diagnostic information including a diagnostic score obtained by scoring the results of reading medical data using a degenerative brain disease diagnostic model within a specific numerical range. Accordingly, the first diagnostic score and the second diagnostic score may have numerical values within a specific range. At this time, the diagnostic score may be scored as close to the largest value or as close to the smallest value within a specific molar range depending on the severity of normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia. For example, the first diagnostic score may have a value between 0 and 1. If the result determined through the degenerative brain disease diagnostic model is close to normal, it has a value close to 0, and if it is close to a degenerative brain disease, it has a value close to 1. It can be created to have a value close to . At this time, depending on the severity of the degenerative brain disease, the first diagnostic score may be generated to have a value closer to 1 when the degenerative brain disease is progressing severely, and a value farther from 1 when the degenerative brain disease is mildly progressing. The first diagnostic score can be generated to have . Additionally, the second diagnostic score may have a value between 0 and 1. For example, if the result determined through the degenerative brain disease diagnosis model is close to Alzheimer's dementia, the second diagnosis score may have a value close to 0, if it is close to Parkinson's disease, it may have a value close to 0.5, and Louis It can have a value close to 1, which is close to corpuscular dementia. Alternatively, the second diagnostic score may represent probability information corresponding to Alzheimer's dementia, Parkinson's disease, or Lewy body dementia, and accordingly, the results determined through the degenerative brain disease diagnostic model are Alzheimer's dementia, Parkinson's disease, and Lewy body dementia. Dementia can have a probability outcome value that adds up to 1.
제1진단모델을 통한 제1진단점수 생성 동작과 제2진단모델을 통한 제2진단 점수를 산출하는 동작은 순차적으로 수행될 수 있다. 예를 들어, 정상과 퇴행성 뇌질환을 판별하기 위해 제1진단모델을 이용하여 제1진단점수를 우선적으로 산출하고, 이어서 퇴행성 뇌질환으로 판별된 입력 데이터를 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 판별하기 위해 제2진단모델을 통해 제2진단점수를 산출할 수 있다.The operation of generating the first diagnostic score through the first diagnostic model and the operation of calculating the second diagnostic score through the second diagnostic model may be performed sequentially. For example, in order to distinguish between normal and degenerative brain disease, the first diagnostic score is first calculated using the first diagnostic model, and then the input data determined as degenerative brain disease is divided into Alzheimer's dementia, Parkinson's disease, and Lewy body dementia. To determine, the second diagnosis score can be calculated through the second diagnosis model.
프로세서 (100)는, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성할 때, 제1진단 모델을 이용하여 제1슬라이스 이미지를 통해 제1진단점수를 산출하고, 제2진단 모델을 이용하여 제2슬라이스 이미지를 통해 제2진단점수를 산출할 수 있다.When the processor 100 generates one or more diagnostic information by inputting input data into a pre-learned degenerative brain disease diagnostic model, the processor 100 calculates a first diagnostic score through the first slice image using the first diagnostic model, Using the second diagnosis model, the second diagnosis score can be calculated through the second slice image.
미리 학습된 퇴행성 뇌질환 진단 모델은 딥 뉴럴 네트워크 (deep neural network; DNN)일 수 있다. 딥 뉴럴 네트워크는 입력 레이어, 출력 레이어를 포함할 수 있고, 또는 입력 레이어 및 출력 레이어 외의 별개의 복수의 히든 레이어를 포함하는 것일 수 있다. 딥 뉴럴 네트워크는 컨볼루셔널 뉴럴 네트워크 (convolutional neural network; CNN), 리커런트 뉴럴 네트워크(recurrent neural network; RNN), 제한 볼츠만 머신 (restricted boltzmann machine; RBM), 심층 신뢰 네트워크(deep belief network; DBN), Q 네트워크, U 네트워크, 샴 네트워크 등을 포함할 수 있고, 예를 들어, 딥 뉴럴 네트워크는 컨볼루셔널 뉴럴 네트워크일 수 있다.The pre-trained degenerative brain disease diagnosis model may be a deep neural network (DNN). A deep neural network may include an input layer and an output layer, or may include a plurality of separate hidden layers other than the input layer and the output layer. Deep neural networks include convolutional neural network (CNN), recurrent neural network (RNN), restricted Boltzmann machine (RBM), and deep belief network (DBN). , may include a Q network, U network, Siamese network, etc., and for example, a deep neural network may be a convolutional neural network.
컨볼루셔널 뉴럴 네트워크는 다계층 퍼셉트론의 한 종류로 컨볼루셔널 레이어를 포함하는 신경망을 포함할 수 있다. 컨볼루셔널 뉴럴 네트워크는 신경망을 통한 연산과정에서 가중치를 이용할 수 있다. 컨볼루셔널 뉴럴 네트워크는 하나 이상의 컨볼루셔널 레이어 및 이와 결합된 신경망 레이어로 구성될 수 있다. 컨볼루셔널 레이어는 필터를 사용하여 입력된 데이터로부터 특징 (feature)를 추출할 수 있다. 이때, 컨볼루셔널 레이어는 필터와 필터를 비선형 값으로 바꾸는 액티베이션 함수를 포함할 수 있다. 컨볼루셔널 뉴럴 네트워크는 이미지 데이터를 차원을 가진 행렬로 나타내어 처리할 수 있으며, 이를 통해 컨볼루셔널 뉴럴 네트워크는 이미지에서 오브젝트를 인식하기 위해 이용될 수 있다. 예를 들어, 레드, 그린 및 블루로 인코딩 된 이미지 데이터는 R, G, B 색상별로 2차원 행렬로 나타내 질 수 있고, 즉, 각 픽셀의 색상 값이 행렬의 성분이 될 수 있으며, 이때, 행렬의 크기는 이미지의 크기와 동일할 수 있다. 컨볼루셔널 뉴럴 네트워크는 풀링 레이어 (Pooling layer)를 포함할 수 있고, 그리고, 이를 통해 2차원 구조의 입력 데이터를 활용하는 것이 가능하다.A convolutional neural network is a type of multi-layer perceptron and may include a neural network including a convolutional layer. A convolutional neural network can use weights in the computational process through a neural network. A convolutional neural network may consist of one or more convolutional layers and neural network layers combined with them. The convolutional layer can extract features from input data using filters. At this time, the convolutional layer may include a filter and an activation function that changes the filter into a non-linear value. A convolutional neural network can process image data by representing it as a matrix with dimensions, and through this, the convolutional neural network can be used to recognize objects in images. For example, image data encoded in red, green, and blue can be represented by R, G, and B colors as a two-dimensional matrix, i.e., the color value of each pixel can be a component of the matrix, where The size of may be the same as the size of the image. A convolutional neural network may include a pooling layer, and through this, it is possible to utilize input data in a two-dimensional structure.
컨볼루셔널 뉴럴 네트워크는 하나 이상의 컨볼루셔널 레이어 및 서브 샘플링 레이어를 포함할 수 있다. 컨벌루셔널 레이어의 출력에는 서브샘플링 레이어가 연결되어 컨벌루셔널 레이어의 출력을 단순화할 수 있다. 예를 들어, 2*2 에버리지 풀링 필터를 가지는 풀링 레이어에 컨벌루셔널 레이어의 출력을 입력시키는 경우, 이미지의 각 픽셀에서 2*2 패치마다 각 패치에 포함되는 평균값을 출력하여 이미지를 압축할 수 있다. 전술한 풀링은 패치에서 최소값을 출력하거나, 패치의 최대값을 출력하는 방식일 수도 있으며 임의의 풀링 방식이 이용될 수 있다. 컨볼루셔널 뉴럴 네트워크는 컨볼루셔널 과정과 풀링 등의 서브 샘플링 과정을 반복적으로 수행하여 주어진 이미지에서 특징 (feature)를 추출할 수 있다. 이때, 컨볼루셔널 레이어 및/또는 서브샘플링 레이어에서 출력은 풀리 커넥티드 레이어 (fully connected layer)에 입력될 수 있다. 풀리 커넥티드 레이어는 하나의 레이어에 있는 모든 뉴런과 이웃한 레이어에 있는 모든 뉴런이 연결되는 레이어이다.A convolutional neural network may include one or more convolutional layers and subsampling layers. A subsampling layer is connected to the output of the convolutional layer to simplify the output of the convolutional layer. For example, when inputting the output of a convolutional layer to a pooling layer with a 2*2 average pooling filter, the image can be compressed by outputting the average value included in each 2*2 patch for each pixel of the image. there is. The above-described pooling may be a method of outputting the minimum value of a patch or the maximum value of a patch, and any pooling method may be used. A convolutional neural network can extract features from a given image by repeatedly performing subsampling processes such as convolutional process and pooling. At this time, the output from the convolutional layer and/or subsampling layer may be input to a fully connected layer. A fully connected layer is a layer in which all neurons in one layer are connected to all neurons in neighboring layers.
프로세서 (100)는, 입력데이터에서 진단 정보를 생성한 원인이 되는 영역을 다른 영역과 구분하여 표시하는 시각 영상을 생성하는 시각화 동작을 수행할 수 있다. 진단 정보에 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수 및 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별을 위한 제2진단점수가 포함되는 경우, 프로세서는 제1진단점수를 생성한 원인이 되는 영역 또는 제2진단점수를 생성한 원인이 되는 영역을 다른 영역과 구분하여 표시하는 시각 영상을 생성하는 시각화 동작을 수행할 수 있다. 시각화 동작은 생성된 진단정보에 기초하여 출력값 및 경사도를 비교하여 입력 데이터에서 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매의 원인이 되는 영역을 표시하고, 이를 다른 영역과 구분하여 표시함으로써 수행될 수 있다. 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매원인이 되는 영역을 다른 영역과 구분하여 표시할 때, 색의 종류 및 밝기를 이용하여 표시할 수 있다. 시각 영상은 영상은 CAM (Class Activation Map) 또는 Grad-CAM이 될 수 있다. The processor 100 may perform a visualization operation to generate a visual image that displays the area responsible for generating diagnostic information in input data by distinguishing it from other areas. If the diagnosis information includes a first diagnosis score for distinguishing between normal and degenerative brain disease and a second diagnosis score for distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia, the processor may determine the cause of generating the first diagnosis score. A visualization operation can be performed to generate a visual image that displays the area or the area causing the second diagnostic score by distinguishing it from other areas. The visualization operation is performed by comparing output values and slopes based on the generated diagnostic information, displaying the areas causing normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia in the input data, and displaying them separately from other areas. You can. When displaying the areas causing normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia separately from other areas, the type and brightness of color can be used to display them. The visual image can be a CAM (Class Activation Map) or Grad-CAM.
프로세서(100)는, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성하는 생성할 때, 진단 정보를 기초로 정상 확률, 알츠하이머성 치매 확률 및 파킨슨병, 루이소체 치매 확률 정보를 포함하는 부분 점수를 산출하는 동작을 수행할 수 있다. 구체적으로, 진단 정보에 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수 및 알츠하이머성 치매와 파킨슨병, 루이소체 치매 판별을 위한 제2진단점수가 포함되는 경우, 프로세서는 제1진단점수 및 상기 제2진단점수를 기초로 정상 확률, 알츠하이머성 치매 확률 및 파킨슨병, 루이소체 치매 확률 정보를 포함하는 부분 점수를 산출하는 동작을 수행할 수 있다. 예를 들어, 프로세서는 제1진단점수를 통해 정상 확률 (a%)을 산출하고, 전체 확률에서 정상 확률을 감산하여 퇴행성 뇌질환 확률 [(1-a)%]을 산출할 수 있다. 그리고, 제2진단 점수를 이용하여 알츠하이머성 치매 확률 (b%)과 전체 확률에서 알츠하이머성 치매 확률을 감산하여 파킨슨병 (c%), 루이소체 치매 확률 [(1-b-c)%]을 산출할 수 있다. 그리고, 이를 종합하여 정상 확률 (a%), 알츠하이머성 치매 확률 [(1-a)*b%] 및 파킨슨병 [(1-a)*(c%)], 루이소체 치매 확률 [(1-a)*(1-b-c)%]을 산출할 수 있다.When the processor 100 generates one or more diagnostic information by inputting input data into a pre-learned degenerative brain disease diagnostic model, the normal probability, Alzheimer's dementia probability, Parkinson's disease, and Lewy body dementia are based on the diagnostic information. An operation may be performed to calculate a partial score including probability information. Specifically, if the diagnostic information includes a first diagnostic score for distinguishing between normal and degenerative brain disease and a second diagnostic score for distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia, the processor Based on the second diagnosis score, an operation can be performed to calculate a partial score including probability information of normalcy, probability of Alzheimer's disease, Parkinson's disease, and probability of Lewy body dementia. For example, the processor may calculate the probability of normalcy (a%) using the first diagnostic score, and calculate the probability of degenerative brain disease [(1-a)%] by subtracting the normal probability from the total probability. Then, using the second diagnosis score, the probability of Parkinson's disease (c%) and Lewy body dementia [(1-b-c)%] can be calculated by subtracting the probability of Alzheimer's dementia from the probability of Alzheimer's disease (b%) and the overall probability. You can. And, combining these, the probability of normalcy (a%), probability of Alzheimer's dementia [(1-a)*b%], Parkinson's disease [(1-a)*(c%)], probability of Lewy body dementia [(1- a)*(1-b-c)%] can be calculated.
프로세서 (100)는, 의료데이터를 미리 학습된 축적값 추정 모델에 입력하여 뇌 영역별 퇴행성 뇌질환 관련 바이오마커의 축적값에 대한 예측값을 산출하는 추정동작을 수행할 수 있다. 구체적으로, 프로세서는 뇌 영역별 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 산출하는 추정동작을 수행할 수 있다.The processor 100 may input medical data into a pre-learned accumulation value estimation model and perform an estimation operation to calculate a predicted value for the accumulation value of a biomarker related to a degenerative brain disease in each brain region. Specifically, the processor may perform an estimation operation to calculate a predicted value for the accumulated value of Alzheimer's dementia-related biomarkers for each brain region.
프로세서 (100)는, 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 산출한 후 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 기초로 진단 정보를 보정하는 동작을 수행할 수 있다. 구체적으로, 프로세서는 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 산출한 후 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 기초로 제2진단 점수를 보정하는 동작을 수행할 수 있다. 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값이 산출되는 경우, 제2진단점수를 통한 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별의 정확도가 더욱 향상될 수 있다. 구체적으로, 알츠하이머성 치매 관련 바이오마커가 아밀로이드인 경우, 아밀로이드 축적의 예측 값이 기준치 이상인 경우 제2진단점수는 알츠하이머성 치매를 나타내는 점수에 가깝게 보정될 수 있고, 아밀로이드가 축적된 예측 값이 기준치 미만인 경우 제2진단점수는 파킨슨병, 루이소체 치매를 나타내는 점수에 더욱 가깝게 보정될 수 있다. 따라서, 알츠하이머성 치매 관련 바이오마커 축적 값을 통한 진단 정보의 보정은 뇌 영역에 축적된 알츠하이머성 치매 관련 바이오마커의 예측값을 추가적으로 진단정보에 반영하기 때문에, 단순히 의료 영상데이터만을 기초로 알츠하이머성 치매와 파킨슨병, 루이소체 치매를 분류하는 과정에 비해 보다 정확하게 알츠하이머성 치매와 파킨슨병, 루이소체 치매를 감별진단하는 것이 가능하다.The processor 100 may calculate a predicted value for an accumulated value of an Alzheimer's dementia-related biomarker and then perform an operation to correct diagnostic information based on the predicted value for an accumulated value of an Alzheimer's dementia-related biomarker. Specifically, the processor may calculate a predicted value for the accumulated value of the Alzheimer's dementia-related biomarker and then perform an operation to correct the second diagnosis score based on the predicted value for the accumulated value of the Alzheimer's dementia-related biomarker. If a predicted value for the accumulated value of Alzheimer's dementia-related biomarkers is calculated, the accuracy of distinguishing Alzheimer's dementia, Parkinson's disease, and Lewy body dementia through the second diagnostic score can be further improved. Specifically, when the Alzheimer's dementia-related biomarker is amyloid, if the predictive value of amyloid accumulation is above the reference value, the second diagnosis score can be corrected to be close to the score indicating Alzheimer's dementia, and if the predictive value of amyloid accumulation is below the reference value In this case, the second diagnosis score can be adjusted to be closer to the score indicating Parkinson's disease or Lewy body dementia. Therefore, the correction of diagnostic information through the accumulated value of Alzheimer's dementia-related biomarkers additionally reflects the predictive value of Alzheimer's dementia-related biomarkers accumulated in the brain region in the diagnostic information, and therefore, the diagnosis and treatment of Alzheimer's dementia based solely on medical imaging data Compared to the process of classifying Parkinson's disease and Lewy body dementia, it is possible to differentially diagnose Alzheimer's disease, Parkinson's disease, and Lewy body dementia more accurately.
이때, 예측값은, 의료데이터에서 수득된 제1대뇌피질 두께 데이터 및 가중치 정보로부터 산출될 수 있다. 가중치 정보는 알츠하이머 관련 바이오마커의 축적 정도를 예측하려는 대상의 대뇌 피질 두께 데이터와 별도의 대뇌 피질 두께 데이터 (이하 "제2대뇌 피질 두께 데이터"로 서술함) 및 제2대뇌 피질 두께 데이터가 획득된 대상과 동일한 대상에서 측정된 바이오마커 축적 데이터 사이에서 정의되는 것일 수 있다. 이때, 제2대뇌 피질 두께 데이터는 자기 공명 영상 (magnetic resonance imaging; MRI)의 T1 강조 영상에서 측정된 것일 수 있고, 바이오마커 축적 데이터는 양전자 방출 단층 촬영 영상 (positron emission tomography; PET)으로부터 수득된 것일 수 있다.At this time, the predicted value can be calculated from first cerebral cortex thickness data and weight information obtained from medical data. The weight information is obtained from the subject's cerebral cortical thickness data for predicting the degree of accumulation of Alzheimer's-related biomarkers, separate cerebral cortical thickness data (hereinafter referred to as "second cerebral cortical thickness data"), and second cerebral cortical thickness data. It may be defined between the subject and the biomarker accumulation data measured in the same subject. At this time, the second cerebral cortex thickness data may be measured from T1-weighted images of magnetic resonance imaging (MRI), and the biomarker accumulation data may be obtained from positron emission tomography (PET) images. It may be.
대뇌피질 두께 데이터는 소프트웨어 프로그램을 이용하여 의료데이터에서 측정될 수 있다. 구체적으로, 대뇌피질 두께 데이터는 소프트웨어를 통해 의료데이터 내의 자기 공명 영상에서 측정될 수 있다. 소프트웨어 프로그램으로는, 예를 들어, FreeSurfer 프로그램이 이용될 수 있다. 따라서, 상기 프로그램을 통해 자기 공명 영상에서 pial/white surface를 생성하여 대뇌피질 부분을 추출하고, 표면 (surface) 위에서 정의된 각 정점 (vertex)의 위치마다 해당하는 대뇌피질 부분의 두께를 구하고, 각 뇌 영역에 해당하는 정점의 평균 두께를 구함에 따라 제1대뇌 피질 두께 데이터가 생성될 수 있다.Cortical thickness data can be measured from medical data using a software program. Specifically, cerebral cortex thickness data can be measured from magnetic resonance images in medical data through software. As a software program, for example, the FreeSurfer program can be used. Therefore, through the above program, a pial/white surface is created from the magnetic resonance image to extract the cerebral cortex portion, the thickness of the corresponding cerebral cortex portion is obtained for each position of each vertex defined on the surface, and each First cerebral cortex thickness data can be generated by calculating the average thickness of the vertex corresponding to the brain region.
이때, 대뇌 피질 두께 데이터는 프로세서에 의해서 의료데이터로부터 생성될 수 있다. 또는, 대뇌 피질 두께 데이터는 별도의 과정으로 의료데이터로부터 수득되고, 프로세서는 단순히 이를 이용하여 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 산출하는 것일 수 있다.At this time, cerebral cortex thickness data may be generated from medical data by a processor. Alternatively, the cerebral cortex thickness data may be obtained from medical data in a separate process, and the processor may simply use this to calculate a predicted value for the accumulated value of biomarkers related to Alzheimer's dementia.
가중치 정보는 각각의 뇌 영역의 제2대뇌 피질 두께와 각각의 뇌 영역의 바이오마커 축적 데이터 사이에서 정의된 것일 수 있다. 이때, 가중치 정보는 일 뇌영역의 알츠하이머성 치매 관련 바이오마커의 축적값과 일 뇌영역과 동일하거나 동일하지 않은 뇌영역의 대뇌 피질 두께 데이터 사이의 연관정도 나타내는 수치값들의 집합일 수 있다. 따라서, 하나의 뇌 영역의 알츠하이머성 치매 관련 바이오마커의 축적값의 예측값을 산출하는 경우, 복수의 가중치들이 이용될 수 있다.Weight information may be defined between the thickness of the second cerebral cortex of each brain region and the biomarker accumulation data of each brain region. At this time, the weight information may be a set of numerical values indicating the degree of correlation between the accumulated value of the Alzheimer's dementia-related biomarker in one brain region and the cerebral cortex thickness data in the brain region that is the same or not the same as the one brain region. Accordingly, when calculating the predicted value of the accumulated value of a biomarker related to Alzheimer's disease in one brain region, a plurality of weights may be used.
예를 들어, 대상에서 획득한 제1대뇌 피질 두께 데이터를 통해 아래쪽 측두이랑 (inferior temporal gyrus; ITG)에 축적된 알츠하이머 관련 바이오마커의 축적 정도를 예측하는 경우를 설명한다. 이 경우, ITG 영역의 제2대뇌 피질 두께 데이터와 ITG 영역의 바이오마커 축적 데이터 사이에서 정의된 가중치 정보와, ITG 영역의 제1대뇌 피질 두께 데이터를 이용하여 ITG 영역에 축적된 알츠하이머 관련 바이오마커의 축적 정도를 예측할 수 있다.For example, a case of predicting the degree of accumulation of Alzheimer's-related biomarkers accumulated in the inferior temporal gyrus (ITG) using first cerebral cortex thickness data obtained from a subject is explained. In this case, weight information defined between the second cerebral cortex thickness data in the ITG region and the biomarker accumulation data in the ITG region, and the first cerebral cortex thickness data in the ITG region are used to determine the Alzheimer's-related biomarkers accumulated in the ITG region. The degree of accumulation can be predicted.
또는, ITG 영역에 축적된 알츠하이머성 치매 관련 바이오마커는, 복수의 뇌영역의 대뇌 피질 두께 데이터와, 상기 복수의 뇌영역에 포함되는 각각의 뇌영역과 ITG 영역 사이의 가중치 정보로부터 예측될 수 있다 (이때, 가중치 정보는 각각의 뇌영역의 제2대뇌 피질 두께 데이터와 ITG 영역의 바이오마커 축적 데이터 사이에서 결정된 것일 수 있음). 이때, 복수의 뇌영역은 경우에 따라 ITG 영역을 포함하지 않을 수 있다. 즉, ITG 영역의 대뇌 피질 두께 및 ITG 영역의 대뇌 피질 두께 및 ITG 영역의 바이오마커 축적 데이터 사이에서 정의된 가중치는 사용되지 않을 수 있다. Alternatively, Alzheimer's dementia-related biomarkers accumulated in the ITG region can be predicted from cerebral cortex thickness data of a plurality of brain regions and weight information between the ITG region and each brain region included in the plurality of brain regions. (At this time, the weight information may be determined between the second cerebral cortex thickness data of each brain region and the biomarker accumulation data of the ITG region). At this time, the plurality of brain regions may not include the ITG region in some cases. That is, the weight defined between the cerebral cortex thickness of the ITG region and the cerebral cortex thickness of the ITG region and the biomarker accumulation data of the ITG region may not be used.
알츠하이머성 치매 관련 바이오마커 축적값에 대한 예측값은, 하나 이상의 뇌영역의 대뇌 피질 두께 데이터와, 상기 하나 이상의 뇌영역에 포함되는 각 뇌영역의 대뇌 피질 데이터와 예측하려는 뇌영역의 바이오마커 축적 데이터 사이에서 정의된 가중치 정보를 이용하여 산출될 수 있다. 구체적으로, 예측하고자 하는 뇌영역의 알츠하이머 관련 바이오마커 축적 값은, 하나 이상의 뇌영역 각각의 대뇌 피질 두께 데이터와 예측하려는 뇌영역의 바이오마커 축적 데이터 사이에서 정의된 가중치와, 하나 이상의 뇌영역에 포함되는 각 뇌영역의 대뇌 피질 두께를 곱한값을 더함으로써 계산될 수 있다.The predicted value for the accumulation value of biomarkers related to Alzheimer's dementia is between the cerebral cortex thickness data of one or more brain regions, the cerebral cortex data of each brain region included in the one or more brain regions, and the biomarker accumulation data of the brain region to be predicted. It can be calculated using the weight information defined in . Specifically, the Alzheimer's-related biomarker accumulation value of the brain region to be predicted is the weight defined between the cerebral cortex thickness data of each of one or more brain regions and the biomarker accumulation data of the brain region to be predicted, and included in one or more brain regions. It can be calculated by adding the product of the cerebral cortex thickness of each brain region.
예를 들어, 알츠하이머 관련 바이오마커의 축적값에 대한 예측값은 하기의 수학식 1에 의해서 계산될 수 있다.For example, the predicted value for the accumulation value of Alzheimer's-related biomarkers can be calculated using Equation 1 below.
[여기서, 는 알츠하이머성 치매 관련 바이오마커 축적 정도 예측 값 (i는 하나의 특정 영역 혹은 전역을 의미할 수 있음), 은 알츠하이머성 치매 관련 바이오마커의 축적값을 예측하고 하는 영역의 수, 는 선정 영역의 대뇌 피질 두께, 는 알츠하이머 관련 바이오마커의 축적값을 예측하고자 하는 영역의 가중치임][here, is the prediction value of the accumulation level of Alzheimer's dementia-related biomarkers (i can refer to one specific area or the entire region), is the number of regions that predict the accumulation value of Alzheimer's dementia-related biomarkers, is the cortical thickness of the selected area, is the weight of the area to predict the accumulation value of Alzheimer's-related biomarkers]
가중치 정보는 머신러닝 모델 또는 딥러닝 모델을 통해 정의될 수 있다. 이때, 머신러닝 모델 또는 딥러닝 모델은 사전 학습된 모델이 이용되거나, 또는, 하나 이상의 제2대뇌 피질 두께 데이터 및 하나 이상의 바이오마커 축적 데이터를 통해 학습과정을 거쳐 확립된 모델이 이용될 수 있다.Weight information can be defined through a machine learning model or deep learning model. At this time, a pre-trained model may be used as the machine learning model or deep learning model, or a model established through a learning process through one or more second cerebral cortex thickness data and one or more biomarker accumulation data may be used.
머신러닝이 이용되는 경우 머신러닝 모델은 의사결정나무 (Decision Tree), 베이지안 망 (Bayesian network) 또는 서포트벡터머신 (support vector machine; SVM) 알고리즘을 포함하는 것일 수 있다. When machine learning is used, the machine learning model may include a decision tree, Bayesian network, or support vector machine (SVM) algorithm.
제2대뇌 피질 두께 데이터 및 같은 대상에서 측정된 바이오마커 축적 데이터는 복수의 쌍일 수 있다. 예를 들어, 제2대뇌 피질 두께 데이터 및 같은 대상에서 측정된 바이오마커 축적 데이터는 10쌍, 20쌍, 30쌍, 50쌍, 100쌍, 200쌍, 500쌍, 1000쌍, 1500쌍, 3000쌍 또는 10,000쌍 이상일 수 있다. 일 실시예에 있어서 가중치는 다수의 제2대뇌 피질 두께 데이터 및 바이오마커 축적 데이터의 연관정도 도출을 통해 정확도가 향상될 수 있다. 이에 따라, 보다 많은 제2대뇌 피질 두께 데이터 및 바이오마커 축적 데이터 쌍을 가중치 정의에 이용하는 경우, 정의되는 가중치 정보의 정확도가 향상될 수 있으며, 이를 통해 가중치 정보를 통한 알츠하이머 관련 바이오마커 축적 예측의 정확도가 향상될 수 있다.The second cerebral cortex thickness data and biomarker accumulation data measured in the same subject may be multiple pairs. For example, secondary cerebral cortex thickness data and biomarker accumulation data measured in the same subject are 10 pairs, 20 pairs, 30 pairs, 50 pairs, 100 pairs, 200 pairs, 500 pairs, 1000 pairs, 1500 pairs, and 3000 pairs. Or it could be more than 10,000 pairs. In one embodiment, the accuracy of the weight may be improved by deriving the degree of correlation between a plurality of second cerebral cortex thickness data and biomarker accumulation data. Accordingly, when more pairs of second cerebral cortex thickness data and biomarker accumulation data are used to define weights, the accuracy of the defined weight information can be improved, and through this, the accuracy of predicting Alzheimer's-related biomarker accumulation through the weight information. can be improved.
일 실시예에 따른 퇴행성 뇌질환 진단 모델은, 하나 이상의 학습영상 및 학습영상에 대응되며 퇴행성 뇌질환 진단 결과를 포함하는 가이드 라벨에 기초하여 지도학습될 수 있다. 지도학습은 퇴행성 뇌질환 진단 모델을 이용하여 학습영상에 대해 생성한 학습 정보와 가이드 라벨의 비교 결과에 기초하여 수행될 수 있다. 지도학습은 학습 정보와 상기 가이드 라벨을 손실함수에 대입하여 계산한 결과값에 기초하여 수행될 수 있다. 예를 들어, 지도학습은 학습 정보와 상기 가이드 라벨을 이진 크로스 엔트로피 (Binary Cross-Entropy) 손실함수에 대입하여 계산한 결과값에 기초하여 수행될 수 있다. 지도학습은, 둘 이상의 학습 데이터를 혼합하는 믹스업 (mix-up) 과정을 통해 조정하여 학습하는 것일 수 있다. 지도학습은 라벨 스무딩 (Label smoothing)을 이용하여 입력된 가이드 라벨과 퇴행성 뇌질환 진단 모델을 이용하여 학습영상에 대해 생성한 학습 정보와의 비교 결과에 기초하여 수행될 수 있다. 구체적으로, 정상과 퇴행성 뇌질환 구분 시 정상을 0, 퇴행성 뇌질환을 1로, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 구분시 알츠하이머성 치매 및 파킨슨병, 루이소체 치매를 확률 값으로로 가이드 라벨을 입력하는 경우를 가정하면, 정상과 퇴행성 뇌질환 구분 시 정상을 0.1, 퇴행성 뇌질환을 0.9로, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매구분시 알츠하이머성 치매를 a 파킨슨병을 b, 루이소체 치매를 c 로 입력(a+b+c=1)하여 학습을 진행할 수 있다.The degenerative brain disease diagnosis model according to one embodiment may be supervised learning based on one or more learning images and a guide label that corresponds to the learning images and includes a degenerative brain disease diagnosis result. Supervised learning can be performed based on the comparison results of learning information and guide labels generated for learning images using a degenerative brain disease diagnosis model. Supervised learning can be performed based on a result calculated by substituting the learning information and the guide label into a loss function. For example, supervised learning can be performed based on a result calculated by substituting the learning information and the guide label into a binary cross-entropy loss function. Supervised learning may be learning by adjusting two or more learning data through a mix-up process. Supervised learning can be performed based on the results of comparison between guide labels input using label smoothing and learning information generated for learning images using a degenerative brain disease diagnosis model. Specifically, when distinguishing between normal and degenerative brain disease, normal is set to 0, degenerative brain disease is set to 1, and when distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia, the guide labels are set as probability values for Alzheimer's dementia, Parkinson's disease, and Lewy body dementia. Assuming that you input , when distinguishing between normal and degenerative brain disease, normal is set to 0.1, degenerative brain disease is set to 0.9, and when distinguishing between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia, Alzheimer's dementia is set to A, Parkinson's disease is set to B, and Lewy body dementia is set to B. You can proceed with learning by entering dementia as c (a+b+c=1).
프로세서 (100)는, 하나 이상의 코어로 구성될 수 있으며, 컴퓨팅 장치의 중앙 처리 장치 (central processing unit; CPU), 그래픽 처리 장치 (graphics processing unit; GPU), 텐서 처리 장치(tensor processing unit; TPU) 등의 데이터 분석, 딥러닝을 위한 프로세서를 포함할 수 있다. 프로세서는, 메모리에 저장된 컴퓨터 프로그램을 판독하여 일 실시예에 따른 기계 학습을 위한 데이터 처리를 수행할 수 있다. 일실시예에 따라 프로세서는, 신경망의 학습을 위한 연산을 수행할 수 있다. 프로세서는, 딥러닝 (deep learning; DL)에서 학습을 위한 입력 데이터의 처리, 입력 데이터에서의 피처 추출, 오차 계산, 역전파 (backpropagation)를 이용한 신경망의 가중치 업데이트 등의 신경망의 학습을 위한 계산을 수행할 수 있다. 프로세서(110)의 CPU, GPU, 및 TPU 중 적어도 하나가 네트워크 함수의 학습을 처리할 수 있다. The processor 100 may be comprised of one or more cores, and may include a central processing unit (CPU), a graphics processing unit (GPU), or a tensor processing unit (TPU) of a computing device. It may include a processor for data analysis and deep learning. The processor may read a computer program stored in a memory and perform data processing for machine learning according to an embodiment. According to one embodiment, the processor may perform calculations for learning a neural network. The processor performs calculations for neural network learning, such as processing input data for learning in deep learning (DL), extracting features from input data, calculating errors, and updating the weights of the neural network using backpropagation. It can be done. At least one of the CPU, GPU, and TPU of the processor 110 may process learning of the network function.
메모리 (200)는, 플래시 메모리 타입 (flash memory type), 하드디스크 타입 (hard disk type), 멀티미디어 카드 마이크로 타입 (multimedia card micro type), 카드 타입의 메모리 (예를 들어 SD 또는 XD 메모리 등), 램 (Random Access Memory; RAM), SRAM (Static Random Access Memory), 롬 (Read-Only Memory; ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), 자기 메모리, 자기 디스크, 광디스크 중 적어도 하나의 타입의 저장매체를 포함할 수 있다.The memory 200 includes a flash memory type, hard disk type, multimedia card micro type, card type memory (for example, SD or XD memory, etc.), RAM (Random Access Memory), SRAM (Static Random Access Memory), ROM (Read-Only Memory; ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, It may include at least one type of storage medium among magnetic disks and optical disks.
일 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하기 위한 동작을 수행하는 컴퓨팅 장치는 필요에 따라 뇌 영상 및 기타 데이터를 송수신하기 위한 네트워크부 및 기타 컴퓨팅 과정에서 이용되는 데이터를 저장 및 관리하기 위한 데이터베이스를 더 포함할 수 있다. 따라서, 일 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하기 위한 동작을 수행하는 컴퓨팅 장치는 MRI 촬영장치, PET 촬영장치 등을 포함하는 뇌촬영 장치들과 유무선 통신이 가능할 수 있다. 또한, 가중치 데이터, 대뇌피질 두께 데이터 및 바이오마커 축적 데이터 등의 데이터는 데이터베이스에 저장될 수 있으며, 데이터베이스에 저장된 데이터는 사용자로부터 인가된 외부 입력에 의해 삭제, 수정 등의 관리가 이루어질 수 있다.According to one embodiment, a computing device that performs an operation to provide information for diagnosing a degenerative brain disease includes a network unit for transmitting and receiving brain images and other data as needed, and storing and managing data used in other computing processes. Additional databases may be included. Accordingly, a computing device that performs an operation to provide information for diagnosing a degenerative brain disease according to an embodiment may be capable of wired or wireless communication with brain imaging devices, including an MRI imaging device and a PET imaging device. Additionally, data such as weight data, cerebral cortex thickness data, and biomarker accumulation data can be stored in a database, and data stored in the database can be managed, such as deletion and modification, by external input authorized by the user.
도 2는 본 발명의 일 실시예에 따른 딥 뉴럴 네트워크를 나타낸 개략도이다.Figure 2 is a schematic diagram showing a deep neural network according to an embodiment of the present invention.
뉴럴 네트워크는 입력 데이터로부터 특징 데이터를 추출하고 특징 데이터를 이용한 추론 (inference) 연산을 제공하도록 설계된 기계 학습 구조의 모델을 나타낼 수 있다. 이때, 특징 데이터는 입력 데이터가 추상화된 특징에 관한 데이터를 나타낼 수 있다. 도 2는 설명의 편의를 위해 히든 레이어가 3개의 레이어를 포함하는 것으로 도시되었으나, 히든 레이어에는 다양한 수의 레이어들을 포함할 수 있다. 신경망은 하나 이상의 레이어를 포함할 수 있으며, 각각의 레이어는 하나 이상의 노드를 포함할 수 있다.A neural network may represent a model of a machine learning structure designed to extract feature data from input data and provide inference operations using feature data. At this time, the feature data may represent data about features in which input data is abstracted. In FIG. 2, for convenience of explanation, the hidden layer is shown as including three layers, but the hidden layer may include a varying number of layers. A neural network may include one or more layers, and each layer may include one or more nodes.
노드 (또는 유닛)은 각 레이어를 구성하는 요소로, 각각의 레이어는 노드 또는 노드들의 집합으로 구성될 수 있다. 뉴럴 네트워크내에서 출력 레이어를 제외한 레이어들의 노드들은 출력 신호를 전송하기 위한 링크들을 통해 다음 레이어의 노드들과 연결될 수 있다. 이때, 각각의 레이어들의 노드들은 링크를 통해 서로 연결될 수 있으며, 연결된 레이어의 노드들은 신호의 송 수신 여부에 따라 입력 노드 및 출력 노드의 관계에 있을 수 있다. 링크를 통해 연결된 노드 중, 출력 노드의 데이터는 입력 노드에 입력된 데이터에 따라 값이 결정될 수 있다. 히든 레이어에 포함된 각각의 노드에는 이전 레이어에 포함된 노드들의 가중된 입력들 (weighted inputs)에 관한 활성 함수 (activation function)의 출력이 입력될 수 있다. 가중된 입력은 이전 레이어에 포함된 노드들의 입력에 가중치(weight)가 반영된 것이다. 이때, 가중치는 가변적일 수 있으며, 신경망의 기능 및 알고리즘에 따라서 가변될 수 있다. 가중치는 뉴럴 네트워크의 파라미터로 지칭될 수 있으며, 활성 함수는 시그모이드 (sigmoid), 하이퍼볼릭 탄젠트 (hyperbolic tangent; tanh) 및 렐루 (rectified linear unit; ReLU)를 포함할 수 있다.A node (or unit) is an element that constitutes each layer, and each layer may be composed of a node or a set of nodes. Within a neural network, nodes in layers other than the output layer can be connected to nodes in the next layer through links for transmitting output signals. At this time, the nodes of each layer may be connected to each other through a link, and the nodes of the connected layers may be in a relationship as an input node and an output node depending on whether signals are transmitted or received. Among nodes connected through a link, the value of the data of the output node may be determined according to the data input to the input node. The output of an activation function regarding the weighted inputs of nodes included in the previous layer may be input to each node included in the hidden layer. Weighted input reflects the weight of the input of nodes included in the previous layer. At this time, the weight may be variable and may vary depending on the function and algorithm of the neural network. Weights may be referred to as parameters of the neural network, and activation functions may include sigmoid, hyperbolic tangent (tanh), and rectified linear unit (ReLU).
최초 입력 노드는 신경망 내의 노드들 중 다른 노드들과의 관계에서 링크를 거치지 않고 데이터가 직접 입력되는 하나 이상의 노드들을 의미할 수 있다. 또는, 신경망 네트워크 내에서, 링크를 기준으로 한 노드 간의 관계에 있어서, 링크로 연결된 다른 입력 노드들을 가지지 않는 노드들을 의미할 수 있다. 이와 유사하게, 최종 출력 노드는 신경망 내의 노드들 중 다른 노드들과의 관계에서, 출력 노드를 가지지 않는 하나 이상의 노드들을 의미할 수 있다. 또한, 히든 노드는 최초 입력 노드 및 최후 출력 노드가 아닌 신경망을 구성하는 노드들을 의미할 수 있다.The initial input node may refer to one or more nodes in the neural network through which data is directly input without going through links in relationships with other nodes. Alternatively, in a neural network network, in the relationship between nodes based on links, it may mean nodes that do not have other input nodes connected by links. Similarly, the final output node may refer to one or more nodes that do not have an output node in their relationship with other nodes among the nodes in the neural network. Additionally, hidden nodes may refer to nodes constituting a neural network other than the first input node and the last output node.
딥 뉴럴 네트워크 (deep neural network; DNN)는 입력 레이어와 출력 레이어 외에 복수의 히든 레이어를 포함하는 신경망을 의미할 수 있다. 딥 뉴럴 네트워크를 이용하면 데이터의 잠재적인 구조 (latent structures)를 파악할 수 있다. 즉, 사진, 글, 비디오, 음성, 음악의 잠재적인 구조 (예를 들어, 어떤 물체가 사진에 있는지, 글의 내용과 감정이 무엇인지, 음성의 내용과 감정이 무엇인지 등)를 파악할 수 있다. 딥 뉴럴 네트워크는 컨볼루션 뉴럴 네트워크 (convolutional neural network; CNN), 리커런트 뉴럴 네트워크 (recurrent neural network; RNN), 오토 인코더 (auto encoder), GAN (Generative Adversarial Networks), 제한 볼츠만 머신 (restricted boltzmann machine; RBM), 심층 신뢰 네트워크 (deep belief network; DBN), Q 네트워크, U 네트워크, 샴 네트워크, 적대적 생성 네트워크(GAN: Generative Adversarial Network) 등을 포함할 수 있다.A deep neural network (DNN) may refer to a neural network that includes a plurality of hidden layers in addition to an input layer and an output layer. Deep neural networks allow you to identify latent structures in data. In other words, it is possible to identify the potential structure of a photo, text, video, voice, or music (e.g., what object is in the photo, what the content and emotion of the text are, what the content and emotion of the voice are, etc.) . Deep neural networks include convolutional neural networks (CNN), recurrent neural networks (RNN), auto encoders, Generative Adversarial Networks (GAN), restricted Boltzmann machines; RBM), deep belief network (DBN), Q network, U network, Siamese network, Generative Adversarial Network (GAN), etc.
컨볼루션 뉴럴 네트워크 (convolutional neural network; CNN)는 동물의 시각 피질의 구성에서 영감을 받아, 이미지와 같은 격자 패턴을 갖는 데이터를 처리하기 위한 딥 러닝 모델의 일종이다. 컨볼루션 뉴럴 네트워크는 일반적으로 컨볼루션 레이어, 풀링 레이어 및 풀리 커넥티드 레이어 (fully connected layer)를 포함할 수 있다. 컨볼루션 레이어 및 풀링 레이어는 신경망 내에 반복적으로 존재할 수 있으며, 입력 데이터는 이러한 레이어 계층을 통해 출력으로 변환될 수 있다. 컨볼루션은 레이어는 특징 추출을 위해서 커널 (또는 마스크)를 이용하여, 커널의 각 요소와 입력 값 간의 요소별 곱은 각각의 위치에서 계산되고 합산되어 출력 값을 얻게되며, 이를 특징 맵 (feature map)이라 지칭한다. 이러한 절차는 임의의 수의 특징 맵을 형성하기 위해 여러 커널을 적용하며 반복될 수 있다. 컨볼루션 뉴럴 네트워크에서 컨볼루션 및 풀링 레이어는 특징 추출을 수행하는 반면, 풀리 커넥티드 레이어는 추출된 특징을 분류하는 동작 등의 최종 출력에 매핑한다. A convolutional neural network (CNN) is a type of deep learning model for processing data with grid patterns such as images, inspired by the organization of the visual cortex of animals. A convolutional neural network may generally include a convolutional layer, a pooling layer, and a fully connected layer. Convolutional layers and pooling layers can exist repeatedly within a neural network, and input data can be transformed into output through these layers. The convolution layer uses a kernel (or mask) to extract features, and the product of each element between each element of the kernel and the input value is calculated and summed at each position to obtain an output value, which is called a feature map. It is referred to as This procedure can be repeated applying multiple kernels to form an arbitrary number of feature maps. In a convolutional neural network, convolutional and pooling layers perform feature extraction, while fully connected layers map the extracted features to the final output, such as a classification operation.
컨볼루션 뉴럴 네트워크는 등의 뉴럴 네트워크는 출력 오류를 최소화하는 방향으로 학습될 수 있다. 입력 레이어에서 출력 레이어로부터 값을 추출하는 순전파과정과 별개로, 뉴럴 네트워크 내에서는 입력된 학습데이터 및 이에 대한 뉴럴 네트워크의 출력값 사이의 오류를 계산하고 이러한 오류를 줄이기 위해 각각의 레이어의 노드들에 대한 가중치를 업데이트 하는 역전파 (backpropagation)가 일어나게 된다. 컨볼루션 뉴럴 네트워크에서 학습시키는 과정은 주어진 학습데이터를 기반으로, 가장 오류가 적은 출력값을 추출하는 커널을 찾는 과정으로 요약될 수 있다. 커널은 컨볼루션 레이어의 훈련과정에서 자동으로 학습되는 유일한 매개변수이다. 반면, 컨볼루션 뉴럴 네트워크에서 커널의 크기, 커널의 수, 패딩 등은 훈련 과정을 시작하기 전에 설정해야하는 하이퍼 파라미터이며, 따라서, 커널의 크기, 커널의 수, 컨볼루션 레이어 및 풀링 레이어의 숫자에 따라서 각기 다른 컨볼루션 뉴럴 네트워크 모델로 구분될 수 있다.Neural networks such as convolutional neural networks can be trained to minimize output errors. Separately from the forward propagation process that extracts values from the input layer to the output layer, within the neural network, the error between the input learning data and the corresponding output value of the neural network is calculated and the nodes of each layer are connected to reduce this error. Backpropagation occurs to update the weights. The learning process in a convolutional neural network can be summarized as the process of finding a kernel that extracts an output value with the fewest errors based on given training data. The kernel is the only parameter that is automatically learned during the training process of the convolutional layer. On the other hand, in convolutional neural networks, the size of the kernel, the number of kernels, padding, etc. are hyperparameters that must be set before starting the training process, and therefore, depending on the size of the kernel, the number of kernels, and the number of convolutional layers and pooling layers, They can be divided into different convolutional neural network models.
뉴럴 네트워크는 각각의 학습 데이터에 정답이 라벨링 된 학습데이터를 이용하는 교사 학습 (supervised learning), 학습 데이터에 정답이 라벨링 되지 않은 비교사 학습 (unsupervised learning), 반교사학습 (semi supervised learning), 또는 강화학습 (reinforcement learning) 중 적어도 하나의 방식으로 학습될 수 있다. 이때, 뉴럴 네트워크를 통한 출력과 라벨 또는 학습 데이터가 비교되면서 오류가 계산될 수 있고, 계산된 오류는 뉴럴 네트워크에서 역방향(즉, 출력 레이어에서 입력 레이어 방향)으로 역전파 되며, 역전파에 따라 뉴럴 네트워크의 각 레이어의 각 노드들의 연결 가중치가 업데이트 될 수 있다. 업데이트 되는 각 노드의 연결 가중치는 학습률(learning rate)에 따라 변화량이 결정될 수 있다.Neural networks use supervised learning using training data in which each training data is labeled with the correct answer, unsupervised learning using training data in which the correct answer is not labeled, semi-supervised learning, or reinforcement. It can be learned in at least one way: reinforcement learning. At this time, the error can be calculated by comparing the output through the neural network and the label or training data, and the calculated error is back-propagated in the neural network in the reverse direction (i.e., from the output layer to the input layer), and the neural network is transmitted according to the back-propagation. The connection weight of each node in each layer of the network may be updated. The amount of change in the connection weight of each updated node may be determined according to the learning rate.
과적합 (overfitting)은 뉴럴 네트워크에서 학습 데이터에 대한 학습이 과하게 일어나게 됨에 따라 학습 수가 증가함에도 오히려 오류가 증가하는 현상이다. 과적합은 머신러닝 알고리즘의 오류를 증가시키는 원인으로 작용할 수 있으며, 이러한 과적합을 막기 위하여 다양한 최적화 방법이 사용될 수 있다. 과적합을 막기 위해서는 학습 데이터를 증가시키거나, 레귤라이제이션 (regularization), 학습의 과정에서 네트워크의 노드 일부를 비활성화하는 드롭아웃 (dropout), 배치 정규화 레이어 (batch normalization layer)의 활용 등의 방법이 적용될 수 있다.Overfitting is a phenomenon in which errors increase even as the number of training increases due to excessive learning on training data in a neural network. Overfitting can cause errors in machine learning algorithms to increase, and various optimization methods can be used to prevent such overfitting. To prevent overfitting, methods such as increasing the learning data, regularization, dropout to disable some of the network nodes during the learning process, and use of a batch normalization layer can be applied. You can.
도 3은 본 발명의 일 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하는 과정을 설명하기 위한 블록구성도를 도시한 도면이다.Figure 3 is a block diagram illustrating a process for providing information for diagnosing a degenerative brain disease according to an embodiment of the present invention.
도 3을 참조하면, 일 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하는 컴퓨팅 장치는 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성 (S101)하고, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성 (S102)하고, 진단 정보를 통해 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 분류 (S103)할 수 있다.Referring to FIG. 3, a computing device that provides information for diagnosing a degenerative brain disease according to an embodiment generates input data based on medical data including a brain image (S101) and converts the input data into a pre-learned degenerative brain disease. One or more diagnostic information can be generated by inputting it into the brain disease diagnosis model (S102), and the diagnostic information can be used to classify normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia (S103).
이때, 컴퓨팅 장치는 하나 이상의 진단 점수를 포함하는 진단 정보를 생성할 수 있다. 구체적으로, 컴퓨팅 장치는 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수 및 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별을 위한 제2진단점수를 포함하는 진단정보를 생성할 수 있다.At this time, the computing device may generate diagnostic information including one or more diagnostic scores. Specifically, the computing device may generate diagnostic information including a first diagnostic score for discriminating between normal and degenerative brain diseases and a second diagnostic score for discriminating between Alzheimer's disease, Parkinson's disease, and Lewy body dementia.
컴퓨팅 장치는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성할 때, 의료데이터를 균일화할 수 있다. 예를 들어, 뇌 자기 공명 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하는 경우, 뇌 자기 공명 영상의 강도 (intensity)를 조절하여 뇌 자기 공명영상을 균일화하여 입력 데이터를 생성할 수 있다. 뇌 자기 공명영상이 균일화되는 경우, 퇴행성 뇌질환 진단 모델의 학습에 유리할 수 있으며, 퇴행성 뇌질환 진단 모델을 통한 진단 정확도가 향상될 수 있다.When generating input data based on medical data including brain images, the computing device may standardize the medical data. For example, when generating input data based on medical data including a brain magnetic resonance image, the input data can be generated by equalizing the brain magnetic resonance image by adjusting the intensity of the brain magnetic resonance image. If brain magnetic resonance images are standardized, it may be advantageous for learning a degenerative brain disease diagnostic model, and diagnostic accuracy through the degenerative brain disease diagnostic model may be improved.
컴퓨팅 장치는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성할 때, 의료데이터에 포함된 영상의 위치를 보정할 수 있다. 예를 들어, 뇌 자기 공명 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하는 경우, 각각의 뇌 자기 공명영상을 기준 템플레이트 (template)로 위치를 맞추어 입력데이터를 생성할 수 있다.When generating input data based on medical data including brain images, the computing device may correct the position of the image included in the medical data. For example, when generating input data based on medical data including a brain magnetic resonance image, the input data can be generated by aligning the position of each brain magnetic resonance image with a reference template.
컴퓨팅 장치는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성할 때, 의료데이터에서 진단과 관련 없는 영역을 제거하는 전처리할 수 있다. 예를 들어, 프로세서는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성할 때, 의료데이터를 두개골 (skull)이 없는 마스크를 이용하여 영상에서 두개골 부분을 제거하는 전처리 과정을 수행할 수 있다.When generating input data based on medical data including brain images, the computing device may preprocess the medical data to remove areas unrelated to diagnosis. For example, when the processor generates input data based on medical data including brain images, it can perform a preprocessing process to remove the skull portion from the image by using a skull-less mask for the medical data. there is.
컴퓨팅 장치는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성할 때, 뇌 영상을 포함하는 의료데이터를 바탕으로 뇌 영역의 슬라이스 이미지를 포함하는 입력데이터를 생성할 수 있다. 예를 들어, 컴퓨팅 장치는 뇌 영상을 포함하는 의료데이터를 기초로 제1슬라이스 이미지 및 제2슬라이스 이미지를 포함하는 입력 데이터를 생성할 수 있다. 이때, 제1슬라이스 이미지 및 제2슬라이스 이미지에는 각기 다른 뇌영역의 이미지가 포함될 수 있다. 구체적으로, 제1슬라이스 이미지는 해마 (hippocampus) 영역 및 해마이랑 (parahippocampal gyrus) 영역의 이미지를 포함할 수 있다. 제2슬라이스 이미지는 전두엽 (frontal lobe) 영역, 뇌섬엽 (insula) 및 측두엽 (temporal lobe) 영역의 이미지를 포함할 수 있다.When generating input data based on medical data including a brain image, the computing device may generate input data including a slice image of a brain region based on the medical data including a brain image. For example, the computing device may generate input data including a first slice image and a second slice image based on medical data including a brain image. At this time, the first slice image and the second slice image may include images of different brain regions. Specifically, the first slice image may include images of the hippocampus region and parahippocampal gyrus region. The second slice image may include images of the frontal lobe area, insula, and temporal lobe area.
컴퓨팅 장치는, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성할 때, 제1슬라이스 이미지를 통해 제1진단점수를 산출하고, 제2슬라이스 이미지를 통해 제2진단점수를 산출할 수 있다. 이때, 전술한 바와 같이, 제1슬라이스 이미지에는 해마 영역 및 해마이랑 영역의 이미지가 포함될 수 있고, 제2슬라이스 이미지에는 전두엽, 뇌섬엽 및 측두엽 영역의 이미지가 포함될 수 있다.When inputting input data into a pre-learned degenerative brain disease diagnosis model to generate one or more diagnostic information, the computing device calculates a first diagnosis score through the first slice image and performs a second diagnosis through the second slice image. Scores can be calculated. At this time, as described above, the first slice image may include images of the hippocampus region and parahippocampal gyrus region, and the second slice image may include images of the frontal lobe, insula, and temporal lobe regions.
일 실시예에서, 미리 학습된 퇴행성 뇌질환 진단 모델은 하나 이상의 진단 모델을 포함하는 것일 수 있다. 예를 들어, 퇴행성 뇌질환 진단 모델은 제1진단점수를 생성하는 제1진단 모델 및 제2진단점수를 생성하는 제2진단 모델을 포함하는 것일 수 있다. 따라서, 컴퓨팅 장치는 하나 이상의 학습된 진단 모델을 통하여 진단 정보를 생성하는 동작을 수행할 수 있다. 이때, 제1진단모델은 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수를 생성하도록 미리 학습된 모델일 수 있다. 제2진단모델은 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별을 위한 제2진단점수를 생성하도록 미리 학습된 모델일 수 있다. In one embodiment, the pre-trained degenerative brain disease diagnostic model may include one or more diagnostic models. For example, a degenerative brain disease diagnostic model may include a first diagnostic model that generates a first diagnostic score and a second diagnostic model that generates a second diagnostic score. Accordingly, the computing device may perform an operation of generating diagnostic information through one or more learned diagnostic models. At this time, the first diagnostic model may be a model trained in advance to generate a first diagnostic score for discriminating between normal and degenerative brain disease. The second diagnosis model may be a model trained in advance to generate a second diagnosis score for distinguishing Alzheimer's disease, Parkinson's disease, and Lewy body dementia.
컴퓨팅 장치는 제1진단모델을 통한 제1진단점수 생성과 제2진단모델을 통한 제2진단 점수를 산출 과정을 순차적으로 수행할 수 있다. 예를 들어, 정상과 퇴행성 뇌질환을 판별하기 위해 제1진단모델을 이용하여 제1진단점수를 우선적으로 산출하고, 이어서 퇴행성 뇌질환으로 판별된 입력 데이터를 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 판별하기 위해 제2진단모델을 통해 제2진단점수를 산출할 수 있다.The computing device may sequentially perform the process of generating a first diagnostic score through a first diagnostic model and calculating a second diagnostic score through a second diagnostic model. For example, in order to distinguish between normal and degenerative brain disease, the first diagnostic score is first calculated using the first diagnostic model, and then the input data determined as degenerative brain disease is divided into Alzheimer's dementia, Parkinson's disease, and Lewy body dementia. To determine, the second diagnosis score can be calculated through the second diagnosis model.
미리 학습된 퇴행성 뇌질환 진단 모델은 딥 뉴럴 네트워크 (deep neural network; DNN)일 수 있다. 딥 뉴럴 네트워크는 입력 레이어, 출력 레이어를 포함할 수 있고, 또는 입력 레이어 및 출력 레이어 외의 별개의 복수의 히든 레이어를 포함하는 것일 수 있다. 딥 뉴럴 네트워크는 컨볼루셔널 뉴럴 네트워크 (convolutional neural network; CNN), 리커런트 뉴럴 네트워크(recurrent neural network; RNN), 제한 볼츠만 머신 (restricted boltzmann machine; RBM), 심층 신뢰 네트워크(deep belief network; DBN), Q 네트워크, U 네트워크, 샴 네트워크 등을 포함할 수 있고, 예를 들어, 딥 뉴럴 네트워크는 컨볼루셔널 뉴럴 네트워크일 수 있다.The pre-trained degenerative brain disease diagnosis model may be a deep neural network (DNN). A deep neural network may include an input layer and an output layer, or may include a plurality of separate hidden layers other than the input layer and the output layer. Deep neural networks include convolutional neural network (CNN), recurrent neural network (RNN), restricted Boltzmann machine (RBM), and deep belief network (DBN). , may include a Q network, U network, Siamese network, etc., and for example, a deep neural network may be a convolutional neural network.
컴퓨팅 장치는, 입력데이터에서 진단 정보를 생성한 원인이 되는 영역을 다른 영역과 구분하여 표시하는 시각 영상을 생성하는 시각화 동작을 수행할 수 있다. 진단 정보에 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수 및 알츠하이머성 치매와 파킨슨병, 루이소체 치매 판별을 위한 제2진단점수가 포함되는 경우, 컴퓨팅 장치는 제1진단점수를 생성한 원인이 되는 영역 또는 제2진단점수를 생성한 원인이 되는 영역을 다른 영역과 구분하여 표시하는 시각 영상을 생성하는 시각화를 수행할 수 있다. 시각화 동작은 생성된 진단정보에 기초하여 출력값 및 경사도를 비교하여 입력 데이터에서 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매의 원인이 되는 영역을 표시하고, 이를 다른 영역과 구분하여 표시함으로써 수행될 수 있다. 정상, 알츠하이머성 치매 및 파킨슨병, 루이소체 치매의 원인이 되는 영역을 다른 영역과 구분하여 표시할 때, 색의 종류 및 밝기를 이용하여 표시할 수 있다. 시각 영상은 영상은 CAM (Class Activation Map) 또는 Grad-CAM이 될 수 있다.The computing device may perform a visualization operation to generate a visual image that displays the area responsible for generating diagnostic information from input data by distinguishing it from other areas. If the diagnostic information includes a first diagnostic score for distinguishing between normal and degenerative brain disease and a second diagnostic score for distinguishing between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia, the computing device determines the cause of generating the first diagnostic score. Visualization can be performed to generate a visual image that displays the area that is affected or the area that causes the second diagnostic score to be distinguished from other areas. The visualization operation is performed by comparing output values and slopes based on the generated diagnostic information, displaying the areas causing normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia in the input data, and displaying them separately from other areas. You can. When displaying the areas causing normal, Alzheimer's, Parkinson's, and Lewy body dementia separately from other areas, the type and brightness of color can be used to display them. The visual image can be a CAM (Class Activation Map) or Grad-CAM.
컴퓨팅 장치는, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성하는 생성할 때, 진단 정보를 기초로 정상 확률, 알츠하이머성 치매 확률 및 파킨슨병, 루이소체 치매 확률 정보를 포함하는 부분 점수를 산출 (S104)할 수 있다. 구체적으로, 진단 정보에 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수 및 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별을 위한 제2진단점수가 포함되는 경우, 컴퓨팅 장치는 제1진단점수 및 상기 제2진단점수를 기초로 정상 확률, 알츠하이머성 치매 확률, 파킨슨병 및 루이소체 치매 확률 정보를 포함하는 부분 점수를 산출할 수 있다.When the computing device generates one or more diagnostic information by inputting input data into a pre-learned degenerative brain disease diagnostic model, normal probability, Alzheimer's dementia probability, Parkinson's disease, and Lewy body dementia probability information are generated based on the diagnostic information. A partial score including can be calculated (S104). Specifically, when the diagnostic information includes a first diagnostic score for distinguishing between normal and degenerative brain disease and a second diagnostic score for distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia, the computing device may store the first diagnostic score and the above. Based on the second diagnosis score, a partial score including information on the probability of normalcy, probability of Alzheimer's disease, Parkinson's disease, and Lewy body dementia can be calculated.
도 4는 본 발명의 다른 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하는 과정을 설명하기 위한 블록구성도를 도시한 도면이다.Figure 4 is a block diagram illustrating a process for providing information for diagnosing a degenerative brain disease according to another embodiment of the present invention.
도 4를 참조하면, 본 발명의 다른 실시예에 따른 퇴행성 뇌질환 진단을 위한 정보를 제공하는 컴퓨팅 장치는, 뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성 (S201)하고, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성 (S202)하고, 진단 정보를 통해 정상, 진단 정보를 보정하기 위해 의료데이터를 미리 학습된 축적값 추정 모델에 입력하여 뇌 영역별 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 산출 (S205)하고, 알츠하이머성 치매 및 파킨슨병, 루이소체 치매를 분류 (S204) 할 수 있다. Referring to FIG. 4, a computing device that provides information for diagnosing a degenerative brain disease according to another embodiment of the present invention generates input data based on medical data including a brain image (S201), and inputs the input data. One or more diagnostic information is generated by inputting it into a pre-learned degenerative brain disease diagnosis model (S202), and in order to correct normal and diagnostic information through the diagnostic information, medical data is input into a pre-learned accumulation value estimation model to determine each brain region. A predicted value for the accumulated value of biomarkers related to Alzheimer's dementia can be calculated (S205), and Alzheimer's dementia, Parkinson's disease, and Lewy body dementia can be classified (S204).
이때, 가중치 정보는 전술한 바와 같이, 각각의 뇌 영역의 제2대뇌 피질 두께와 각각의 뇌 영역의 바이오마커 축적 데이터 사이에서 정의된 것일 수 있다. 이때, 가중치 정보는 일 뇌영역의 알츠하이머성 치매 관련 바이오마커의 축적값과 일 뇌영역과 동일하거나 동일하지 않은 뇌영역의 대뇌 피질 두께 데이터 사이의 연관정도 나타내는 수치값들의 집합일 수 있다. 따라서, 하나의 뇌 영역의 알츠하이머성 치매 관련 바이오마커의 축적값의 예측값을 산출하는 경우, 복수의 가중치들이 이용될 수 있다.At this time, as described above, the weight information may be defined between the thickness of the second cerebral cortex of each brain region and the biomarker accumulation data of each brain region. At this time, the weight information may be a set of numerical values indicating the degree of correlation between the accumulated value of the Alzheimer's dementia-related biomarker in one brain region and the cerebral cortex thickness data in the brain region that is the same or not the same as the one brain region. Accordingly, when calculating the predicted value of the accumulated value of a biomarker related to Alzheimer's disease in one brain region, a plurality of weights may be used.
알츠하이머성 치매 관련 바이오마커 축적값에 대한 예측값은, 하나 이상의 뇌영역의 대뇌 피질 두께 데이터와, 상기 하나 이상의 뇌영역에 포함되는 각 뇌영역의 대뇌 피질 데이터와 예측하려는 뇌영역의 바이오마커 축적 데이터 사이에서 정의된 가중치 정보를 이용하여 산출될 수 있다. 구체적으로, 예측하고자 하는 뇌영역의 알츠하이머 관련 바이오마커 축적 값은, 하나 이상의 뇌영역 각각의 대뇌 피질 두께 데이터와 예측하려는 뇌영역의 바이오마커 축적 데이터 사이에서 정의된 가중치와, 하나 이상의 뇌영역에 포함되는 각 뇌영역의 대뇌 피질 두께를 곱한값을 더함으로써 계산될 수 있다.The predicted value for the accumulation value of biomarkers related to Alzheimer's dementia is between the cerebral cortex thickness data of one or more brain regions, the cerebral cortex data of each brain region included in the one or more brain regions, and the biomarker accumulation data of the brain region to be predicted. It can be calculated using the weight information defined in . Specifically, the Alzheimer's-related biomarker accumulation value of the brain region to be predicted is the weight defined between the cerebral cortex thickness data of each of one or more brain regions and the biomarker accumulation data of the brain region to be predicted, and included in one or more brain regions. It can be calculated by adding the product of the cerebral cortex thickness of each brain region.
예를 들어, 알츠하이머 관련 바이오마커의 축적값에 대한 예측값은 하기의 수학식 2에 의해서 계산될 수 있다.For example, the predicted value for the accumulation value of Alzheimer's-related biomarkers can be calculated using Equation 2 below.
[여기서, 는 알츠하이머성 치매 관련 바이오마커 축적 정도 예측 값 (i는 하나의 특정 영역 혹은 전역을 의미할 수 있음), 은 알츠하이머성 치매 관련 바이오마커의 축적값을 예측하고 하는 영역의 수, 는 선정 영역의 대뇌 피질 두께, 는 알츠하이머 관련 바이오마커의 축적값을 예측하고자 하는 영역의 가중치임][here, is the prediction value of the accumulation level of Alzheimer's dementia-related biomarkers (i can refer to one specific area or the entire region), is the number of regions that predict the accumulation value of Alzheimer's dementia-related biomarkers, is the cortical thickness of the selected area, is the weight of the area to predict the accumulation value of Alzheimer's-related biomarkers]
컴퓨팅 장치는, 산출된 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 통해 진단 정보를 보정 (S203)할 수 있다. 구체적으로, 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 통해 알츠하이머성 치매와 파킨슨병, 루이소체 치매 판별을 위한 제2진단점수를 보정할 수 있다. 예를 들어, 컴퓨팅 장치는 예측값이 기준치 이상인 경우, 제2진단점수를 알츠하이머성 치매를 나타내는 점수에 보다 가깝게 보정할 수 있다. 또는, 컴퓨팅 장치는 예측값이 기준치 이하인 경우, 제2진단점수를 파킨슨병 및 루이소체 치매를 나타내는 점수에 가깝게 보정할 수 있다.The computing device may correct the diagnostic information through the calculated predicted value of the accumulated value of the Alzheimer's dementia-related biomarker (S203). Specifically, the second diagnostic score for distinguishing between Alzheimer's dementia, Parkinson's disease, and Lewy body dementia can be corrected through the predicted value of the accumulated value of biomarkers related to Alzheimer's dementia. For example, if the predicted value is above the reference value, the computing device may correct the second diagnosis score to be closer to the score indicating Alzheimer's dementia. Alternatively, if the predicted value is below the reference value, the computing device may correct the second diagnosis score to be closer to the score representing Parkinson's disease and Lewy body dementia.
이하, 본 발명을 하기의 실시예에 의하여 더욱 상세히 설명한다. 그러나 이들 실시예는 본 발명을 예시하기 위한 것일 뿐이며, 본 발명의 범위가 이들 실시예에 의하여 한정되는 것은 아니다.Hereinafter, the present invention will be described in more detail through the following examples. However, these examples are only for illustrating the present invention, and the scope of the present invention is not limited by these examples.
실시예 1: 딥러닝 기반의 정상, 알츠하이머성 치매 및 파킨슨병, 루이소체 치매의 감별진단Example 1: Differential diagnosis of normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia based on deep learning
ADNI 및 동아대에서 제공받은 자기 공명 영상을 이용하여 하기의 실험을 진행하였다.The following experiment was conducted using magnetic resonance images provided by ADNI and Dong-A University.
획득한 자기 공명 영상을 균일화하기 위하여 N4 Bias Field Correction 진행하여, 각각의 자기 공명 영상의 강도 (intensity)를 조절하여 균일하게 맞추어 주었다. 이후, 각 자기 공명 영상의 위치를 보정하기 위해 ANTS (Advanced Normalization Tools) 적용하였으며, 기준 템플레이트 (template)로 위치를 맞추어 주었다.In order to uniformize the obtained magnetic resonance images, N4 Bias Field Correction was performed, and the intensity of each magnetic resonance image was adjusted to make it uniform. Afterwards, ANTS (Advanced Normalization Tools) was applied to correct the position of each magnetic resonance image, and the position was adjusted to the reference template.
이어서, ITK-snap 소프트웨어를 이용하여 자기 공명 영상에서 뇌 영역의 슬라이스 이미지를 추출하였다. 이때, 정상과 퇴행성 뇌질환을 구분하기 위한 슬라이스 이미지는 axial 방향으로 z-119 축을 기준으로 한 단면을, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 분류하기 위한 슬라이스 이미지는 coronal 방향으로 y-140 축을 기준으로 한 단면을 이용하였다. 이후, 추출된 이미지를 두개골(skull)이 없는 mask를 이용하여 두개골(skull) 부분을 제거하였으며, 이미지에서 뇌 이미지가 있는 부분을 사용하기 위해서 검은 배경 부분을 잘라내었다. 딥러닝 모델에 넣기 위해 모델의 입력값에 맞도록 이미지의 크기를 변경하였으며, 이미지의 분포를 균일하게 하기 위해 정규화 (normalization)를 진행하였다.Subsequently, slice images of brain regions were extracted from the magnetic resonance images using ITK-snap software. At this time, the slice image for distinguishing between normal and degenerative brain diseases is a cross section based on the z-119 axis in the axial direction, and the slice image for classifying Alzheimer's dementia, Parkinson's disease, and Lewy body dementia is a cross section based on the z-119 axis in the coronal direction. A cross section based on the axis was used. Afterwards, the skull part of the extracted image was removed using a mask without the skull, and the black background part was cut out to use the part of the image with the brain image. To be included in the deep learning model, the size of the image was changed to fit the model's input value, and normalization was performed to make the distribution of the image uniform.
이어서, 진단 모델의 학습을 진행하였다. 사용한 모델은 Inception ResNet v2. (CNN classification model)였으며, 해당 모델은 299x299x3의 이미지를 받아 35x35x256의 특징 (feature)를 뽑아내고 Inception A, B, C를 거쳐 맨 마지막에는 sigmoid (binary classification의 경우)를 거치는 구조이다. 학습을 위해 전처리된 이미지를 랜덤한 이미지 두 개를 합쳐 넣는 방식인 믹스-업을 이용하여 입력하였다. 가이드 라벨 입력 시 라벨 스무딩 (Label smoothing)을 이용하여, 0 대신에 0.1, 1 대신에 0.9 값을 입력하였다. 이때, 정상인과 퇴행성 뇌질환 분류는 정상인이 0.1 퇴행성 뇌질환이 0.9를 입력하였다. Learning rate는 0.025를 사용하였고 총 5개의 fold를 이용하여 학습을 진행하였고, 손실함수는 이진 크로스 엔트로피 (Binary Cross-Entropy) 손실함수를 사용하였다. Next, learning of the diagnostic model was conducted. The model used is Inception ResNet v2. (CNN classification model), the model receives an image of 299x299x3, extracts features of 35x35x256, goes through Inception A, B, C, and finally goes through sigmoid (in the case of binary classification). For learning, preprocessed images were input using mix-up, which is a method of combining two random images. When entering the guide label, label smoothing was used and values of 0.1 instead of 0 and 0.9 instead of 1 were entered. At this time, the classification between normal people and degenerative brain disease was 0.1 for normal people and 0.9 for degenerative brain disease. The learning rate was 0.025, learning was conducted using a total of 5 folds, and the loss function was the Binary Cross-Entropy loss function.
진단 모델의 학습 후, 학습하지 않은 이미지들에 대해 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 판별하였다. 이때, 학습된 모델이 어느 부분을 보고 각각 정상과 알츠하이머성 치매, 알츠하이머성 치매와 파킨슨병, 루이소체 치매를 분류했는지 확인하기 위하여 Grad-CAM 알고리즘을 사용하였으며, 정상과 퇴행성 뇌질환 판별의 경우 정상 판단의 근거가 되는 영역은 파란색으로, 퇴행성 뇌질환의 판별의 근거가 되는 영역은 붉은 색으로 나타내었다. 전술한 과정에 의해 판별한 결과를 도 5 내지 도 6 및 표 1 내지에 나타내었다.After learning the diagnostic model, normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia were identified for the untrained images. At this time, the Grad-CAM algorithm was used to check which parts the learned model looked at and classified normal and Alzheimer's disease, Alzheimer's dementia and Parkinson's disease and Lewy body dementia, respectively. In the case of normal and degenerative brain disease discrimination, normal The area that serves as the basis for judgment is shown in blue, and the area that serves as the basis for discrimination of degenerative brain disease is shown in red. The results determined by the above-described process are shown in Figures 5 and 6 and Table 1.
RMRI - Center RMRI-Center |
평가방법 (Evaluation method)Evaluation method method) |
정상 (CN)Normal (CN) | 퇴행성 뇌질환degenerative brain disease | 정확도 (ACC) %Accuracy (ACC) % |
ADNIADNI | EnsembleEnsemble | 447447 | 288288 | 85.1785.17 |
판별결과, 도 5 및 표 1에서 확인할 수 있듯이, 일 실시예에 따른 방법은 퇴행성 뇌질환과 정상을 판별할 수 있음을 확인하였다. As can be seen from the determination results in FIG. 5 and Table 1, it was confirmed that the method according to one embodiment can distinguish between degenerative brain disease and normal brain disease.
실시예 2: 진단 점수와 인지기능검사 점수 사이의 상관관계 확인Example 2: Confirmation of correlation between diagnosis score and cognitive function test score
ADNI에서 제공받은, 인지기능검사 점수(MMSE, CDR-SB)와 이에 대응하는 자기 공명 영상을 이용하여 하기의 실험을 진행하였다.The following experiment was conducted using cognitive function test scores (MMSE, CDR-SB) and corresponding magnetic resonance images provided by ADNI.
획득한 자기 공명 영상을 상기 실시예 1과 같이 처리한 뒤, 실시예 1에서 획득한 모델들을 이용해 ensemble method로 자기 공명 영상의 퇴행성 뇌질환 진단 예측 점수를 구하였다. 이후, 구한 퇴행성 뇌질환 진단 예측 점수와 인지기능검사 점수 간의 상관 관계를 계산하였다.After processing the obtained magnetic resonance image as in Example 1, the prediction score for degenerative brain disease diagnosis from the magnetic resonance image was obtained using the ensemble method using the models obtained in Example 1. Afterwards, the correlation between the obtained degenerative brain disease diagnosis prediction score and the cognitive function test score was calculated.
실험결과, 도 7에서 확인할 수 있듯이, ADNI cohort에서는 CDR-SB에 대해서 0.7612, MMSE에 대해서 -0.7435으로 각각 강한 양적/음적 선형 관계를 보였다. 따라서, 본 발명에 따른 방법은 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 감별 진단할 뿐만 아니라 생성하는 진단 점수를 통해 퇴행성 뇌질환이 진행된 경우 퇴행성 뇌질환 진행 정도까지 함께 추정할 수 있음을 확인하였다.As can be seen from the experimental results in Figure 7, the ADNI cohort showed a strong positive/negative linear relationship of 0.7612 for CDR-SB and -0.7435 for MMSE, respectively. Therefore, the method according to the present invention not only differentially diagnoses normal, Alzheimer's dementia, Parkinson's disease, and Lewy body dementia, but also can estimate the degree of progression of the degenerative brain disease when the degenerative brain disease has progressed through the generated diagnostic score. Confirmed.
본 발명은 자기공명 영상의 딥러닝 판독 기술에 기반한 다양한 퇴행성 뇌질환 감별진단 방법 및 장치에 관한 것으로, 더욱 상세하게는, 딥러닝 기술을 이용하여 자기공명 영상 기반으로 정상 (Cognitive Normal; CN), 알츠하이머성 치매 (Alzheimer's Disease; AD), 파킨슨병 (Parkinson's Disease; PD) 및 루이소체 치매 (Dementia with Lewy bodies; DLB)를 감별진단하는 방법 및 장치에 관한 것이다.The present invention relates to a method and device for differential diagnosis of various degenerative brain diseases based on deep learning interpretation technology of magnetic resonance images. More specifically, the present invention relates to a method and device for differential diagnosis of cognitive normal (CN), cognitive normal (CN) based on magnetic resonance images using deep learning technology. It relates to a method and device for differential diagnosis of Alzheimer's Disease (AD), Parkinson's Disease (PD), and Dementia with Lewy bodies (DLB).
Claims (14)
- 컴퓨팅 장치에 의해 의료 영상데이터에서 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법에 있어서, In a method of providing information for diagnosing a degenerative brain disease from medical image data by a computing device,뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하는 처리 단계;A processing step of generating input data based on medical data including brain images;상기 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성하는 생성 단계; 및A generation step of generating one or more diagnostic information by inputting the input data into a pre-learned degenerative brain disease diagnostic model; and상기 진단 정보를 통해 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 분류하는 분류 단계;A classification step of classifying normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia based on the diagnostic information;를 포함하고, Including,상기 진단 정보는 정상과 퇴행성 뇌질환 판별을 위한 제1진단점수 및 알츠하이머성 치매, 파킨슨병 및 루이소체 치매 판별을 위한 제2진단점수를 포함하는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.The diagnostic information provides information for diagnosing degenerative brain diseases, including a first diagnostic score for distinguishing between normal and degenerative brain diseases and a second diagnostic score for distinguishing between Alzheimer's disease, Parkinson's disease, and Lewy body dementia. method.
- 제1항에 있어서, 상기 퇴행성 뇌질환 진단 모델은,The method of claim 1, wherein the degenerative brain disease diagnostic model is:상기 제1진단점수를 생성하는 제1진단 모델 및 상기 제2진단점수를 생성하는 제2진단 모델을 포함하는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A method of providing information for diagnosing a degenerative brain disease, comprising a first diagnostic model for generating the first diagnostic score and a second diagnostic model for generating the second diagnostic score.
- 제1항에 있어서, 상기 의료데이터는,The method of claim 1, wherein the medical data is:뇌 자기공명영상 (magnetic resonance imaging; MRI) 데이터를 포함하는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A method of providing information for diagnosing a degenerative brain disease, comprising brain magnetic resonance imaging (MRI) data.
- 제1항에 있어서, 상기 입력 데이터는 제1슬라이스 이미지 및 제2슬라이스 이미지를 포함하고,The method of claim 1, wherein the input data includes a first slice image and a second slice image,상기 제1슬라이스 이미지는 해마 (hippocampus) 영역 및 해마이랑 (parahippocampal gyrus) 영역의 이미지를 포함하고,The first slice image includes images of the hippocampus region and the parahippocampal gyrus region,상기 제2슬라이스 이미지는 전두엽 (frontal lobe) 영역, 뇌섬엽 (insula) 및 측두엽 (temporal lobe) 영역의 이미지를 포함하는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A method of providing information for diagnosing a degenerative brain disease, wherein the second slice image includes images of the frontal lobe region, insula, and temporal lobe region.
- 제4항에 있어서, According to paragraph 4,상기 제1진단점수는 상기 제1슬라이스 이미지를 통해 산출되는 것이고, 상기 제2진단점수는 상기 제2슬라이스 이미지를 통해 산출되는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A method of providing information for diagnosing a degenerative brain disease, wherein the first diagnostic score is calculated through the first slice image, and the second diagnostic score is calculated through the second slice image.
- 제1항에 있어서, 상기 생성 단계는The method of claim 1, wherein the generating step is상기 제1진단점수 및 상기 제2진단점수를 기초로 정상 확률, 알츠하이머성 치매 확률, 파킨슨병 확률 및 루이소체 치매 확률 정보를 포함하는 부분 점수를 산출하는 산출 단계;를 포함하는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A calculation step of calculating a partial score including normal probability, Alzheimer's dementia probability, Parkinson's disease probability, and Lewy body dementia probability information based on the first diagnosis score and the second diagnosis score; a degenerative brain comprising a. A method of providing information for diagnosing a disease.
- 제1항에 있어서, 상기 방법은,The method of claim 1, wherein입력데이터에서 제1진단점수를 생성한 원인이 되는 영역 또는 제2진단점수를 생성한 원인이 되는 영역을 다른 영역과 구분하여 표시하는 시각 영상을 생성하는 시각화 단계;를 포함하는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A visualization step of generating a visual image that displays the area causing the first diagnostic score or the area causing the second diagnostic score in the input data by distinguishing it from other areas; a degenerative brain comprising a. A method of providing information for diagnosing a disease.
- 제1항에 있어서, 상기 퇴행성 뇌질환 진단 모델은,The method of claim 1, wherein the degenerative brain disease diagnostic model is:하나 이상의 학습영상 및 상기 학습영상에 대응되며 퇴행성 뇌질환 진단 결과를 포함하는 가이드 라벨에 기초하여 지도학습되며,Supervised learning is performed based on one or more learning images and guide labels that correspond to the learning images and include diagnostic results for degenerative brain diseases,상기 지도학습은 퇴행성 뇌질환 진단 모델을 이용하여 학습영상에 대해 생성한 학습 정보와 가이드 라벨의 비교 결과에 기초하여 수행되는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A method of providing information for diagnosing a degenerative brain disease, wherein the supervised learning is performed based on a comparison result between learning information generated for a learning image and a guide label using a degenerative brain disease diagnosis model.
- 제8항에 있어서, 상기 지도학습은, The method of claim 8, wherein the supervised learning is:상기 학습 정보와 상기 가이드 라벨을 이진 크로스 엔트로피 (Binary Cross-Entropy) 손실함수에 대입하여 계산한 결과값에 기초하여 수행되는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A method of providing information for diagnosing a degenerative brain disease, which is performed based on a result calculated by substituting the learning information and the guide label into a binary cross-entropy loss function.
- 제1항에 있어서, 상기 방법은,The method of claim 1, wherein상기 의료데이터를 미리 학습된 축적값 추정 모델에 입력하여 뇌 영역별 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 산출하는 추정 단계; 및An estimation step of inputting the medical data into a previously learned accumulation value estimation model to calculate a predicted value for the accumulation value of Alzheimer's dementia-related biomarkers for each brain region; and상기 알츠하이머성 치매 관련 바이오마커의 축적값에 대한 예측값을 기초로 상기 진단 정보를 보정하는 보정 단계를 포함하는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A method of providing information for diagnosing a degenerative brain disease, comprising a correction step of correcting the diagnostic information based on a predicted value of the accumulated value of the Alzheimer's dementia-related biomarker.
- 제10항에 있어서, 상기 예측값은,The method of claim 10, wherein the predicted value is:상기 의료데이터에서 수득된 제1대뇌피질 두께 데이터 및 가중치 정보로부터 산출되고, Calculated from the first cerebral cortex thickness data and weight information obtained from the medical data,상기 가중치 정보는 하나 이상의 제2대뇌 피질 두께 데이터와 하나 이상의 바이오마커 축적 데이터 사이에서 정의되는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A method of providing information for diagnosing a degenerative brain disease, wherein the weight information is defined between one or more second cerebral cortex thickness data and one or more biomarker accumulation data.
- 제8항에 있어서, 상기 지도학습은,The method of claim 8, wherein the supervised learning is:둘 이상의 학습 데이터를 혼합하는 믹스업 (mix-up) 과정을 통해 조정하여 학습하는 것인, 퇴행성 뇌질환 진단을 위한 정보를 제공하는 방법.A method of providing information for diagnosing degenerative brain diseases by adjusting and learning through a mix-up process that mixes two or more learning data.
- 저장 매체에 저장된 컴퓨터 프로그램으로서, A computer program stored on a storage medium, comprising:컴퓨터 프로그램은 하나 이상의 프로세서에서 실행되는 경우, If a computer program runs on more than one processor,퇴행성 뇌질환 진단을 위한 정보를 제공하기 위한 이하의 동작들을 수행하도록 하며, Perform the following actions to provide information for diagnosing degenerative brain diseases,상기 동작들은: The above operations are:뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하는 처리 동작; A processing operation that generates input data based on medical data including brain images;입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성하는 생성 동작; A generation operation of generating one or more diagnostic information by inputting input data into a pre-learned degenerative brain disease diagnostic model;및 진단 정보를 통해 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 분류하는 분류 동작;and a classification operation to classify normal, Alzheimer's disease, Parkinson's disease, and Lewy body dementia through diagnostic information;을 포함하는 것인, 저장 매체에 저장된 컴퓨터 프로그램.A computer program stored on a storage medium that includes a.
- 퇴행성 뇌질환 진단을 위한 정보를 제공하기 위한 컴퓨팅 장치로서, A computing device for providing information for diagnosing degenerative brain diseases,하나 이상의 코어를 포함하는 프로세서; A processor including one or more cores;및 메모리;를 포함하고,and memory;프로세서는, The processor is뇌 영상을 포함하는 의료데이터를 기초로 입력 데이터를 생성하고, 입력 데이터를 미리 학습된 퇴행성 뇌질환 진단 모델에 입력하여 하나 이상의 진단 정보를 생성하고, 및 진단 정보를 통해 정상, 알츠하이머성 치매, 파킨슨병 및 루이소체 치매를 분류하는, 컴퓨팅장치.Input data is generated based on medical data including brain images, the input data is input into a pre-trained degenerative brain disease diagnosis model to generate one or more diagnostic information, and the diagnostic information is used to generate normal, Alzheimer's dementia, and Parkinson's diseases. A computing device that classifies diseases and Lewy body dementia.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020220044444A KR20230145724A (en) | 2022-04-11 | 2022-04-11 | Methods and apparatus for differential diagnosis of various degenerate brain disease based on deep learning technology of Magnet Resonance image classification |
KR10-2022-0044444 | 2022-04-11 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023200182A1 true WO2023200182A1 (en) | 2023-10-19 |
Family
ID=88329936
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/KR2023/004666 WO2023200182A1 (en) | 2022-04-11 | 2023-04-06 | Method and device for differentially diagnosing various degenerative brain diseases on basis of deep-learning reading technology in magnetic resonance imaging |
Country Status (2)
Country | Link |
---|---|
KR (1) | KR20230145724A (en) |
WO (1) | WO2023200182A1 (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101696061B1 (en) * | 2015-03-05 | 2017-01-13 | 재단법인 아산사회복지재단 | Methods and apparatus for acquiring image based biomarkers suitable for diagnosis of neurodegenerative diseases and methods for diagnosing neurodegenerative diseases |
KR101943011B1 (en) * | 2018-01-22 | 2019-01-28 | 주식회사 뷰노 | Method for facilitating medical image reading and apparatus using the same |
KR20210096292A (en) * | 2019-03-06 | 2021-08-04 | 주식회사 뉴로젠 | Brain disease prediction device and method, learning device for predicting brain disease |
KR102321601B1 (en) * | 2020-11-19 | 2021-11-05 | 주식회사 휴런 | Biological Classification Devices and Methods for Alzheimer's Disease Using Multimodal Brain Image |
-
2022
- 2022-04-11 KR KR1020220044444A patent/KR20230145724A/en not_active Application Discontinuation
-
2023
- 2023-04-06 WO PCT/KR2023/004666 patent/WO2023200182A1/en unknown
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101696061B1 (en) * | 2015-03-05 | 2017-01-13 | 재단법인 아산사회복지재단 | Methods and apparatus for acquiring image based biomarkers suitable for diagnosis of neurodegenerative diseases and methods for diagnosing neurodegenerative diseases |
KR101943011B1 (en) * | 2018-01-22 | 2019-01-28 | 주식회사 뷰노 | Method for facilitating medical image reading and apparatus using the same |
KR20210096292A (en) * | 2019-03-06 | 2021-08-04 | 주식회사 뉴로젠 | Brain disease prediction device and method, learning device for predicting brain disease |
KR102321601B1 (en) * | 2020-11-19 | 2021-11-05 | 주식회사 휴런 | Biological Classification Devices and Methods for Alzheimer's Disease Using Multimodal Brain Image |
Non-Patent Citations (1)
Title |
---|
ALISSA MOHAMAD; LONES MICHAEL A.; COSGROVE JEREMY; ALTY JANE E.; JAMIESON STUART; SMITH STEPHEN L.; VALLEJO MARTA: "Parkinson’s disease diagnosis using convolutional neural networks and figure-copying tasks", NEURAL COMPUTING AND APPLICATIONS, SPRINGER LONDON, LONDON, vol. 34, no. 2, 8 September 2021 (2021-09-08), London, pages 1433 - 1453, XP037672236, ISSN: 0941-0643, DOI: 10.1007/s00521-021-06469-7 * |
Also Published As
Publication number | Publication date |
---|---|
KR20230145724A (en) | 2023-10-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109447183B (en) | Prediction model training method, device, equipment and medium | |
Sharma et al. | A deep learning based convolutional neural network model with VGG16 feature extractor for the detection of Alzheimer Disease using MRI scans | |
Al-Adhaileh | Diagnosis and classification of Alzheimer's disease by using a convolution neural network algorithm | |
WO2023095989A1 (en) | Method and device for analyzing multimodality medical images for cerebral disease diagnosis | |
Alroobaea et al. | Alzheimer's Disease Early Detection Using Machine Learning Techniques | |
WO2021261808A1 (en) | Method for displaying lesion reading result | |
Alkabawi et al. | Computer-aided classification of multi-types of dementia via convolutional neural networks | |
WO2023173538A1 (en) | Alzheimer disease evaluation method and system, and device and storage medium | |
WO2023249217A1 (en) | Methodology and pipeline for extracting multi-modal brain-imaging-based brain disease biomarker | |
WO2023200182A1 (en) | Method and device for differentially diagnosing various degenerative brain diseases on basis of deep-learning reading technology in magnetic resonance imaging | |
Baskar et al. | An Accurate Prediction and Diagnosis of Alzheimer’s Disease using Deep Learning | |
Gao et al. | Brain Age Prediction with 3D ResNet34 Model in Healthy Control, Mild Cognitive Impairment, and Alzheimer's Disease | |
Piñeros-Fernández | Artificial intelligence applications in the diagnosis of neuromuscular diseases: a narrative review | |
Praveen et al. | Prediction of Alzheimer's Disease using Deep Learning Algorithms | |
WO2023204496A1 (en) | High-speed technique and apparatus for quantifying cerebral cortex shrinkage on basis of deep learning | |
Amulya et al. | Classification of brain images for Alzheimer's disease detection | |
KR20230128210A (en) | Device and method of medical image regression for the diagnosis of dementia | |
Arabi et al. | High Accuracy Diagnosis for MRI Imaging Of Alzheimer’s Disease using Xgboost | |
Varma et al. | A Short Review on Automatic Detection of Glaucoma Using Fundus Image | |
WO2023153839A1 (en) | Dementia information calculation method and analysis device using two-dimensional mri | |
Seshadri et al. | Early prediction of Alzheimer’s disease with a multimodal multitask deep learning model | |
Amin et al. | Automated psoriasis detection using deep learning | |
WO2024090677A1 (en) | Method for calculating dementia-related index on basis of volume of extracerebral cerebrospinal fluid, and analysis apparatus | |
Devi et al. | A Hybrid Approach to detect and characterize Alzheimer’s Disease using Robust PCA and Random Forest Algorithm | |
Singh et al. | Alzheimer's Dementia and its Various Biomarker Details with Several Detection Methodologies-A Comprehensive Review |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 23788533 Country of ref document: EP Kind code of ref document: A1 |