WO2022168969A1 - 学習装置、学習済みモデルの生成方法、診断処理装置、コンピュータプログラム及び診断処理方法 - Google Patents

学習装置、学習済みモデルの生成方法、診断処理装置、コンピュータプログラム及び診断処理方法 Download PDF

Info

Publication number
WO2022168969A1
WO2022168969A1 PCT/JP2022/004607 JP2022004607W WO2022168969A1 WO 2022168969 A1 WO2022168969 A1 WO 2022168969A1 JP 2022004607 W JP2022004607 W JP 2022004607W WO 2022168969 A1 WO2022168969 A1 WO 2022168969A1
Authority
WO
WIPO (PCT)
Prior art keywords
brain
sbr
patient
image data
neural network
Prior art date
Application number
PCT/JP2022/004607
Other languages
English (en)
French (fr)
Japanese (ja)
Inventor
一尋 井汲
宇植 史
Original Assignee
株式会社Medicolab
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 株式会社Medicolab filed Critical 株式会社Medicolab
Priority to JP2022579638A priority Critical patent/JPWO2022168969A1/ja
Publication of WO2022168969A1 publication Critical patent/WO2022168969A1/ja

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B10/00—Instruments for taking body samples for diagnostic purposes; Other methods or instruments for diagnosis, e.g. for vaccination diagnosis, sex determination or ovulation-period determination; Throat striking implements
    • 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
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • 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
    • 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

Definitions

  • the present disclosure relates to a learning device, a method of generating a trained model, a diagnostic processing device, a computer program, and a diagnostic processing method.
  • Isotope testing is a technique for visualizing the physiological function of a target site by intravenously injecting a testing agent containing a radioactive isotope. Using this technology, it is possible to determine whether physiological function is normal or abnormal, which cannot be obtained by general imaging tests.
  • the disease differentiation support device of Patent Document 1 uses the image data of a nuclear medicine examination that images the head of a subject to whom a dopamine transporter imaging agent is administered into the body. It assists in distinguishing between a degenerative first disease and a second disease that is different from the first disease.
  • isotope testing has several problems.
  • hospitals must order special radiological examination agents each time a patient undergoes an examination, requiring complicated and time-consuming management.
  • it is labor intensive because specialized examination staff such as radiological technologists are required, and each examination takes a long time.
  • Fifth, the examination cost per test is high, which places a heavy financial burden on the patient.
  • an object of the present invention is to provide a learning device, a method for generating a trained model, a diagnostic processing device, a computer program, and a diagnostic processing method that place less burden on patients and hospitals.
  • the degree of degeneration of dopamine neurons in the nigrostriatal cortex which corresponds to the SBR (Specific Binding Ratio) value of brain dopamine transporter scintigraphy examination, is determined by deep learning. It is a predictive analysis.
  • the learning device of the present invention stores image data of head MRI examinations of the heads of a plurality of patients and SBR values of brain dopamine transporter scintigraphy examinations corresponding to each patient. and an SBR value of cerebral dopamine transporter scintigraphy examination corresponding to each patient stored in the storage unit, using the image data of head MRI examination of a plurality of patients stored in the storage unit as input data. and a learning unit that generates a trained model using deep learning as teacher data.
  • the trained model includes a first neural network that uses right-brain images to predictively analyze right-brain SBR values, and a second neural network that uses left-brain images to predictively analyze left-brain SBR values.
  • the trained model uses right brain images and left brain images and includes a third neural network that predictively analyzes right brain SBR values and left brain SBR values.
  • the trained model includes a fourth neural network that predicts and analyzes right-brain SBR values and left-brain SBR values using horizontally-reversed right-brain images or left-brain images.
  • the method of generating a trained model of the present invention comprises a storage step of storing image data of a head MRI examination in which the heads of a plurality of patients are photographed, and SBR values of a cerebral dopamine transporter scintigraphy examination corresponding to each patient. using the image data of head MRI examinations of a plurality of patients stored in the storage step as input data, and the SBR values of brain dopamine transporter scintigraphy examinations corresponding to the respective patients stored in the storage unit as teaching data; generating a model using deep learning.
  • the diagnostic processing apparatus of the present invention has an input unit for inputting image data of a patient's head MRI examination, and a head MRI examination image data obtained by photographing the heads of a plurality of patients as input data, and corresponding to each patient. Based on the image data of the patient's head MRI examination input from the input unit, the above-mentioned a predictive analysis unit for predictively analyzing the SBR value of the patient's brain dopamine transporter scintigraphy test.
  • the trained model includes a first neural network that uses right-brain images to predictively analyze right-brain SBR values, and a second neural network that uses left-brain images to predictively analyze left-brain SBR values.
  • the trained model uses right brain images and left brain images and includes a third neural network that predictively analyzes right brain SBR values and left brain SBR values.
  • the trained model includes a fourth neural network that predicts and analyzes right-brain SBR values and left-brain SBR values using horizontally-reversed right-brain images or left-brain images.
  • It has a display unit that displays the normal range of the SBR and the predicted and analyzed SBR value.
  • the present invention is a computer program to be executed by a diagnostic processing apparatus. SBR value of brain dopamine transporter scintigraphy test of the input patient based on input image data of head MRI test of the patient, using a trained model generated using deep learning as teacher data. and a predictive analysis step of predictively analyzing the above.
  • the present invention is a diagnostic processing method performed by a diagnostic processing apparatus, comprising: an input step of inputting image data of a head MRI examination of a patient's head; and using a trained model generated using deep learning with the SBR value of brain dopamine transporter scintigraphy examination corresponding to each patient as teacher data, head MRI examination of the patient input in the input step and a predictive analysis step of predictively analyzing the SBR value of the input patient's brain dopamine transporter scintigraphy test based on the image data of the above.
  • a learning device a method of generating a trained model, a diagnostic processing device, a computer program, and a diagnostic processing method that place less burden on patients and medical institutions.
  • FIG. 4 is an explanatory diagram of specific preprocessing of image data for head MRI examination;
  • FIG. 10 schematically shows processing such as resizing and halving of 10 intermediate DICOM images included in a sample of image data of head MRI examination.
  • FIG. 4 is a table showing average errors and average squared errors for each of the three models created for each of the neural networks NN1 to NN4;
  • FIG. 2 is an explanatory diagram of a neural network used in an embodiment of the present invention;
  • FIG. 10 is a diagram for explaining SBR prediction analysis results of the right brain without normalization processing in the neural network NN1;
  • FIG. 10 is a diagram for explaining SBR prediction analysis results of the right brain when normalization processing is performed in the neural network NN1;
  • FIG. 10 is a diagram for explaining SBR prediction analysis results of the left brain without normalization processing in the neural network NN2;
  • FIG. 10 is a diagram for explaining SBR prediction analysis results of the left brain when normalization processing is performed in the neural network NN2;
  • FIG. 10 is a diagram for explaining SBR prediction analysis results of the right brain and left brain without normalization processing in the neural network NN3;
  • FIG. 10 is a diagram for explaining SBR prediction analysis results for the right brain and left brain when normalization processing is performed in the neural network NN3;
  • FIG. 10 is a diagram for explaining SBR prediction analysis results of the right brain and left brain without normalization processing in the neural network NN4;
  • FIG. 10 is a diagram for explaining SBR prediction analysis results of the right brain and left brain when normalization processing is performed in the neural network NN4;
  • Software that predicts and analyzes the SBR value of brain dopamine transporter scintigraphy examination from the image data of head MRI examination.
  • Screen that selects a folder containing DICOM image files to be analyzed and imports all files in it. For example. It is an example of a display screen of all DICOM images contained in a folder. It is an example of the display screen of the SBR prediction analysis value displayed on the display part of the diagnostic processing apparatus in embodiment of this invention.
  • FIG. 1 is a diagram for explaining the outline of the present invention.
  • the present invention uses quantitative values obtained from isotope examinations as teacher data and data obtained from specific examination modalities (image data and quantitative data along time series) as input data.
  • This software is equipped with a trained model generated using deep learning as a predictive analysis algorithm.
  • This invention aims to limit the number of patients who truly need isotope testing or to replace isotope testing by obtaining predictive analysis results comparable to isotope testing from less burdensome testing modalities.
  • the examination modalities for obtaining input data may be not only a single modality but also a combination of multiple modalities.
  • Metadata such as data acquisition conditions for input data and teacher data is also included in the predictive analysis algorithm. For example, imaging conditions, image conditions, and protocol conditions such as postural changes and exercise loads are included.
  • FIG. 2 shows an example of predictive analysis of brain dopamine transporter scintigraphy results from head MRI examination image data.
  • Cerebral dopamine transporter scintigraphy involves injecting radioactive iodine-labeled isoflurane intravenously, and imaging the isoflurane bound to dopamine transporters (DAT) in dopamine nerve endings in the nigrostriatal body with a gamma camera. This is an imaging test that visualizes the density and distribution of
  • Cerebral dopamine transporter scintigraphy is a test performed in patients with suspected Parkinson's syndrome and is used to differentiate essential tremor or drug-induced Parkinsonism from other Parkinson's syndromes. This test is normal in essential tremor and drug-induced parkinsonism, whereas it is abnormal in other parkinsonisms.
  • An SBR value is output as a quantitative value.
  • ROI region of interest
  • the SBR value is calculated from the total striatal area count, the count per unit volume of the whole brain excluding the striatum, the volume of the striatal area, and the actual volume of the striatum. From this, the degree of degeneration and loss can be determined, since it reflects the distribution, density, and function of dopamine-producing neurons.
  • a neural network is generated for predictive analysis of brain dopamine transporter scintigraphy SBR values from head MRI examination image data.
  • Imaging sequences for head MRI include T1-weighted images, T2-weighted images, FLAIR images (fluid-attenuated inversion recovery), diffusion-weighted images, T2 * images, SWI (susceptibility-weighted imaging) images, melanin-weighted images, MRA (MR-angiography) images, etc.
  • File formats of image data include DICOM, TIFF, JPG, and the like.
  • Imaging conditions include the MRI manufacturer name, model name, magnetic field strength, coil name (number of channels), repetition time (TR) for each sequence, echo time (TE), inversion time (TI), slice thickness, Field of View, Matrix Size, Parallel Imaging/Acceleration factor, Band Width, Number of Excitation, Scan Time, etc.
  • SBR values supervised data for brain dopamine transporter scintigraphy are standardized values among facilities.
  • Imaging conditions include gamma camera maker name, model name, collimator name, drug maker name, dose at test time, actual dose, energy, matrix size, magnification, time from administration to photography, start to end including the time of
  • the data collection mode is a continuous rotation collection mode, it also includes the number of seconds per phase, the rotation angle, the number of views, and the like. If the data collection mode is the Step and Shoot mode, the Step Angle, the number of seconds required for one step, the Step and Shoot time, etc. are also included.
  • FIG. 3(a) shows a head MRI examination, (b) brain dopamine transporter scintigraphy, (c) input data, and (d) teacher data.
  • All image data e.g., DICOM images
  • one type of imaging sequence e.g., FLAIR images
  • a neural network was developed using samples from multiple patients as input data and brain dopamine transporter scintigraphy data (Right SBR value/Left SBR value) corresponding to each patient as training data.
  • FIG. 4 shows an example of predictive analysis of the SBR value of the patient's brain dopamine transporter scintigraphy examination from the image data of the patient's head MRI examination.
  • the diagnostic processing apparatus 100 includes a learning unit 10, a learned model 11, a storage unit 12, an input unit 13, a prediction analysis unit 14, a display unit 15, a storage unit 16, have
  • the input unit 13 inputs image data (test MRI data) D15 of the patient's head MRI examination.
  • the learning unit 10 performs deep learning using the head MRIDICOM data D11 stored in the storage unit 12 as input data and the SBR value D12 of the brain dopamine transporter scintigraphy test of the patient stored in the storage unit 12 as teacher data. generates a trained model 11.
  • the trained model 11 includes a first neural network NN1, a second neural network NN2, a third neural network NN3 and a fourth neural network NN4, which will be described later.
  • the predictive analysis unit 14 predictively analyzes the patient's SBR value based on the patient's head MRI examination image data D15 input from the input unit 13 using the learned model 11 (step S12). At this time, the patient's age is also entered for comparison with the normal range of SBR, which varies with age.
  • the display unit 15 displays various screens of the diagnostic processing device 100 based on the data stored in the storage unit 16, and also displays the normal range of SBR and the predicted and analyzed SBR value.
  • Input data and teacher data were collected from one or more hospitals, and these data sets were used to generate a supervised learning neural network NN. At that time, 90% of the data set was used as the training data set D13, and the remaining 10% was used as the verification data set D14.
  • the model 10a being learned, while continuing the learning of the model using the training data set D13 and the verification data set D14, the already trained model 10 is used, and the test MRI data input to the input unit 13 From D15, the predictive analysis unit 14 predictively analyzed the SBR (step S12), and evaluated the learning result of the model (step S13). This time, the test MRI data D15 is substituted with the same data as the verification MRI data D14.
  • the storage unit 12 stores image data (head MRIDICOM data) D11 of head MRI examination obtained by photographing the heads of a plurality of patients as input data, and teacher data for each patient.
  • the corresponding brain dopamine transporter scintigraphy SBR value (dopamine transporter scintigraphy data) D12 is stored.
  • the storage unit 12 also stores image data of head MRI examinations of multiple patients as input data, and metadata such as data acquisition conditions of brain dopamine transporter scintigraphy examinations corresponding to each patient as teacher data. included.
  • FIG. 5 is an explanatory diagram of specific preprocessing of image data for head MRI examination.
  • FIG. 6 schematically shows processing such as resizing and halving of 10 intermediate DICOM images included in a sample of image data of head MRI examination.
  • One sample is all image data (e.g., DICOM image) from the cranial side to the caudal side included in one type of imaging sequence (e.g., FLAIR image) obtained in one head MRI examination of one patient. point (step S21).
  • image data e.g., DICOM image
  • one type of imaging sequence e.g., FLAIR image
  • the images were divided in half (step S23) to obtain a left half (right brain) image D21 of 224 ⁇ 224 pixels and a right half (left brain) image D22 of 224 ⁇ 224 pixels (step S24).
  • the left half (right brain) image D21 of 224 ⁇ 224 pixels was inverted to obtain ten 224 ⁇ 224 pixel inverted left half (right brain) images D23 (step S25).
  • the left half (right brain) 224 ⁇ 224 pixel image D21 which is one of the halved images, is inverted to increase data homogeneity with the right half (left brain) image D22. gone.
  • the first neural network NN1 to the fourth neural network NN4 were trained using the images of a plurality of patients as input data and the SBR values of brain dopamine transporter scintigraphy corresponding to each patient as training data.
  • the first neural network NN1 is a neural network that uses left half (right brain) images and predictively analyzes right brain SBR values.
  • the second neural network NN2 is a neural network that uses right half (left brain) images and predictively analyzes left brain SBR values.
  • the third neural network NN3 is a neural network that uses left half (right brain) images and right half (left brain) images to predict and analyze right brain SBR values and left brain SBR values, respectively.
  • the fourth neural network NN4 is a neural network that predicts and analyzes the SBR value of the right brain and the SBR value of the left brain by using the left half (right brain) image and the right half (left brain) image that are horizontally inverted.
  • FIG. 7 is a diagram for explaining normalization processing of signal intensities between samples.
  • FIG. 7(a) is the average signal intensity of each sample before normalization, and (b) is the maximum signal intensity of each sample before normalization.
  • FIG. 7(c) shows the average signal intensity after normalization processing of each sample, and (d) shows the average signal intensity and the maximum signal intensity after normalization processing of each sample.
  • FIG. 8 is a table showing the mean error and the mean squared error for each of the three models created for each of the first neural network NN1 to the fourth neural network NN4. For each of the first neural network NN1 to the fourth neural network NN4, three models were created each with and without normalization processing of signal strength between samples as additional preprocessing.
  • FIG. 9 is an explanatory diagram of the neural network used in the embodiment of the present invention.
  • 3D Convolution Layer indicates a three-dimensional convolution layer.
  • 3D Batch Normalization Layer indicates a three-dimensional normalization processing layer.
  • ReLU Layer indicates a normalization processing linear function and is a kind of activation function.
  • Flatten F1 transforms the shape of the data into a vector of [1,K].
  • 3D ConvBNActiBlock 50 is a layer block in which 3D Convolutional Layer 51, 3D Batch Normalization 52 and ReLU Layer 53 are combined in this order.
  • a 3D ConvActiBlock 60 is a layer block in which a 3D Convolutional Layer 61 and a ReLU Layer 62 are combined in this order.
  • 10 preprocessed DICOM images of size 224 ⁇ 224 are input data, and the first 3D ConvBNActiBlock 50a (kernel size is 3 ⁇ 3 ⁇ 3 (depth ⁇ height ⁇ width) stride). is 1 x 2 x 2 (depth x height x width), padding is 1 x 1 x 1 (depth x height x width), and 1 x 10 x 224 x 224 to 64 x Convert to 10x112x112.
  • the second 3D ConvBNActiBlock 50b (kernel size is 3x3x3, stride is 2x2x2, no padding) goes from 64x10x112x112 to 128x4x55 Convert to x55.
  • a third 3D ConvBNActiBlock 50c (kernel size is 3x3x3; stride is 3x3x3; no padding), from 128x4x55x55 to 512x1x18 Convert to x18.
  • Fourth 3D ConvBNActiBlock 50d (kernel size is 1x3x3, stride is 3x3x3, no padding), from 512x1x18x18 to 1024x1x6 Convert to x6.
  • one 3DConvActiBlock 60 (kernel size is 1x6x6, stride is 1x1x1, no padding), from 1024x1x6x6 to 1024x1x1 Convert to x1. Convert to a 1 ⁇ 1024 vector with flatten F1.
  • the 1024 ⁇ 512 fully connected layer 70a converts from 1 ⁇ 1024 to 1 ⁇ 512.
  • a 512 ⁇ 128 fully connected layer 70b converts from 1 ⁇ 512 to 1 ⁇ 128.
  • the 128x1 fully connected layer 70c outputs the value of SBR as a 1x1 scalar.
  • the optimizer uses Adam, and the weight of the L2 regularization term is 0.3.
  • the loss function used the mean squared error (MSE) loss function.
  • the number of data samples is 284 training data samples and 32 verification (test) data samples for the first neural network 1NN and the second neural network NN2 that predict and analyze only one SBR value.
  • the number of training data samples was 567 and the number of validation (test) data samples was 63 for the third neural network NN3 and the fourth neural network NN4 that predictively analyze both SBR values. Note that one sample includes 10 intermediate DICOM images.
  • the batch size for training and validation (test) data is 2.
  • the learning rates are 0.0008, 0.002, 0.003, 0.00001, 0.00008, 0.0001, 0.0002, 0.0003, 0.0004, 0.0005, 0.0006, 0. 0007, 0.0009, 0.001, 0.005, 0.008, 0.01.
  • the development environment is Windows (registered trademark) 10, and the framework used is PyTorch.
  • FIG. 10 is a diagram for explaining SBR prediction analysis results of the right brain without normalization processing in the first neural network NN1.
  • FIG. 10(a) is a diagram showing the distribution of the predicted value and the true value by the first neural network NN1
  • (b) is a diagram showing the number of error samples between the predicted value and the true value
  • (c) is a diagram showing the distribution of errors by age.
  • the normal value of the SBR value may differ depending on the age, so we are also looking at the variation according to the age.
  • the mean error was 1.079211 and the mean squared error was 2.002554.
  • FIG. 11 is a diagram for explaining SBR prediction analysis of the right brain with normalization processing in the first neural network NN1.
  • FIG. 11(a) is a diagram showing the distribution of the predicted value and the true value by the first neural network NN1
  • (b) is a diagram showing the number of error samples between the predicted value and the true value
  • (c) is a diagram showing the distribution of errors by age.
  • the mean error was 1.033450 and the mean squared error was 2.359399.
  • FIG. 12 is a diagram for explaining the SBR prediction analysis result of the left brain without normalization processing in the second neural network NN2.
  • FIG. 12(a) is a diagram showing the distribution of the predicted value and the true value by the second neural network NN2
  • (b) is a diagram showing the number of error samples between the predicted value and the true value
  • (c) is a diagram showing the distribution of errors by age.
  • the mean error was 0.958857 and the mean squared error was 1.446806.
  • FIG. 13 is a diagram for explaining the SBR prediction analysis result of the left brain when normalization processing is performed in the second neural network NN2.
  • FIG. 13(a) is a diagram showing the distribution of the predicted value and the true value by the second neural network NN2
  • (b) is a diagram showing the number of error samples between the predicted value and the true value
  • (c) is a diagram showing the distribution of errors by age.
  • the mean error was 0.925819 and the mean squared error was 1.501469.
  • FIG. 14 is a diagram for explaining SBR prediction analysis results for the right brain and left brain without normalization processing in the third neural network NN3.
  • FIG. 14(a) is a diagram showing the distribution of the predicted value and the true value by the third neural network NN3
  • (b) is a diagram showing the number of error samples between the predicted value and the true value
  • (c) is a diagram showing the distribution of errors by age.
  • the mean error was 1.495315 and the mean squared error was 3.332895.
  • FIG. 15 is a diagram for explaining SBR prediction analysis results for the right brain and left brain when normalization processing is performed in the third neural network NN3.
  • FIG. 15(a) is a diagram showing the distribution of the predicted value and the true value by the neural network NN3,
  • (b) is a diagram showing the number of error samples between the predicted value and the true value, and
  • (c) is FIG. 10 is a diagram showing the distribution of error by age; The mean error was 1.635853 and the mean squared error was 3.769612.
  • FIG. 16 is a diagram for explaining SBR prediction analysis results for the right brain and left brain without normalization processing in the fourth neural network NN4.
  • FIG. 16(a) is a diagram showing the distribution of the predicted value and the true value by the fourth neural network NN4, (b) is a diagram showing the number of error samples between the predicted value and the true value, and (c) is a diagram showing the distribution of errors by age.
  • the mean error was 1.273068 and the mean squared error was 2.794731.
  • FIG. 17 is a diagram for explaining SBR prediction analysis results for the right brain and left brain when normalization processing is performed in the fourth neural network NN4.
  • FIG. 17(a) is a diagram showing the distribution of the predicted value and the true value by the fourth neural network NN4,
  • (b) is a diagram showing the number of error samples between the predicted value and the true value,
  • (c) is a diagram showing the distribution of errors by age.
  • the mean error was 1.441400 and the mean squared error was 3.066445.
  • FIG. 18 is a screen example of software for predicting and analyzing the SBR value of brain dopamine transporter scintigraphy examination from the image data of head MRI examination. As shown in FIG. 18, first, press the file import button on the upper left and select the folder in which the sample is stored, or drag and drop the folder to the left frame.
  • FIG. 19 is a display example of all DICOM images contained in the folder. As shown in FIG. 19, when the analysis button on the lower right is pressed, predictive analysis of the SBR value of the brain dopamine transporter scintigraphy test is started from the image data of the head MRI test by pressing the analyze button. .
  • FIG. 20 is a diagram showing a display example of the SBR prediction value displayed on the display unit of the diagnostic processing device according to the embodiment of the present invention.
  • the predicted and analyzed SBR(R) value and SBR(L) value are displayed in the drawing, and numerical values are also displayed in the lower frame.
  • the lower SBR(R) values are indicated by dark circles and the upper SBR(L) values are indicated by lighter circles.
  • the SBR value also decreases with normal aging, and the solid line in the center in FIG. 2 x standard deviation). Therefore, the region between the upper dashed line and the lower dashed line is the normal range, indicating that both the SBR(R) and SBR(L) values predicted and analyzed are decreased.
  • the normal range for SBR is one that decreases with age.
  • the SBR (R) value and SBR (L) value can be predicted and analyzed from the image data of the patient's head MRI examination, and the error between the predicted value and the true value varies depending on age, but is small. There are many types of data, and the function as a diagnostic processing device is implemented by comparing the predicted value and the normal range. This is clear from the performance evaluation results of the trained model 11 created in 1.
  • the diagnostic processing apparatus 100 functions as a method for generating the trained model 11, and the method for generating the trained model 11 is based on the image data of the head MRI examination in which the heads of a plurality of patients are photographed and the brain corresponding to each patient.
  • the storage step of storing the SBR value of the dopamine transporter scintigraphy test in the storage unit 12, and the image data of the head MRI examination of the heads of a plurality of patients stored in the storage step by the learning unit 10 are used as input data, and generating a trained model 11 generated by deep learning using SBR values of a brain dopamine transporter scintigraphy test corresponding to each patient as teacher data.
  • image data of head MRI examinations of multiple patients which is input data
  • data acquisition conditions of cerebral dopamine transporter scintigraphy examinations corresponding to each patient which is teacher data.
  • Metadata may also be part of the input data.
  • the diagnostic processing apparatus 100 has a computer program executed on the diagnostic processing apparatus 100.
  • This computer program receives image data of head MRI examinations of a plurality of patients as input data, and generates brain dopamine transduction data corresponding to each patient.
  • the input is the image data of the patient's head MRI test, and the brain dopamine transporter scintigraphy test of the patient.
  • the diagnostic processing device 100 is caused to execute a predictive analysis step of predictively analyzing the SBR value, a display step of displaying the normal range of SBR, and the predicted and analyzed SBR value.
  • the diagnostic processing method performed by the diagnostic processing apparatus 100 uses image data of head MRI examinations of a plurality of patients as input data, and deep learning using SBR values of brain dopamine transporter scintigraphy examinations corresponding to each patient as teacher data.
  • a predictive analysis step of predictively analyzing the SBR value of the brain dopamine transporter scintigraphy test of the patient using the input as the image data of the patient's head MRI test using the trained model 11 generated in and
  • a display step of displaying the normal range and the predicted analyzed SBR value is displaying the normal range and the predicted analyzed SBR value.
  • the trained model 11 uses right brain images, predictively analyzes right brain SBR values with the first neural network NN1, and uses left brain images, predictively analyzes left brain SBR values with the second neural network NN2.
  • the trained model 11 uses the right brain image and the left brain image, and predicts and analyzes the SBR value of the right brain and the SBR value of the left brain by the third neural network NN3.
  • the trained model 11 uses a left-right inverted right brain image or left brain image, and predicts and analyzes the right brain SBR value and the left brain SBR value by the fourth neural network NN4.
  • the computer programs and software that operate on the diagnostic processing device 100 are read from the storage unit and executed by the CPU to realize various functions.
  • diagnostic processing device 10 learning unit 11 trained model 12 storage unit 13 input unit 14 prediction analysis unit 15 display unit 16 storage unit NN1 first neural network NN2 second neural network NN3 third neural network NN4 fourth neural network

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Public Health (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Pathology (AREA)
  • Veterinary Medicine (AREA)
  • Epidemiology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Primary Health Care (AREA)
  • Radiology & Medical Imaging (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)
PCT/JP2022/004607 2021-02-05 2022-02-07 学習装置、学習済みモデルの生成方法、診断処理装置、コンピュータプログラム及び診断処理方法 WO2022168969A1 (ja)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2022579638A JPWO2022168969A1 (enrdf_load_stackoverflow) 2021-02-05 2022-02-07

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2021-017014 2021-02-05
JP2021017014 2021-02-05

Publications (1)

Publication Number Publication Date
WO2022168969A1 true WO2022168969A1 (ja) 2022-08-11

Family

ID=82741174

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2022/004607 WO2022168969A1 (ja) 2021-02-05 2022-02-07 学習装置、学習済みモデルの生成方法、診断処理装置、コンピュータプログラム及び診断処理方法

Country Status (2)

Country Link
JP (1) JPWO2022168969A1 (enrdf_load_stackoverflow)
WO (1) WO2022168969A1 (enrdf_load_stackoverflow)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2019146936A (ja) * 2018-02-28 2019-09-05 富士フイルム株式会社 診断支援システム、診断支援方法、及びプログラム
US20200027557A1 (en) * 2018-02-28 2020-01-23 Human Longevity, Inc. Multimodal modeling systems and methods for predicting and managing dementia risk for individuals
WO2020054803A1 (ja) * 2018-09-12 2020-03-19 株式会社Ｓｐｌｉｎｋ 診断支援システムおよび方法
JP2020166809A (ja) * 2019-03-28 2020-10-08 キヤノンメディカルシステムズ株式会社 モデル学習用のシステム、装置、及び学習方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2019146936A (ja) * 2018-02-28 2019-09-05 富士フイルム株式会社 診断支援システム、診断支援方法、及びプログラム
US20200027557A1 (en) * 2018-02-28 2020-01-23 Human Longevity, Inc. Multimodal modeling systems and methods for predicting and managing dementia risk for individuals
WO2020054803A1 (ja) * 2018-09-12 2020-03-19 株式会社Ｓｐｌｉｎｋ 診断支援システムおよび方法
JP2020166809A (ja) * 2019-03-28 2020-10-08 キヤノンメディカルシステムズ株式会社 モデル学習用のシステム、装置、及び学習方法

Also Published As

Publication number Publication date
JPWO2022168969A1 (enrdf_load_stackoverflow) 2022-08-11

Similar Documents

Publication Publication Date Title
JP7357927B2 (ja) 診断支援システムおよび方法
Bui et al. Medical imaging informatics
CN111247595B (zh) 用于诊断认知疾病病理和/或结果的医疗系统
Maclaren et al. Reliability of brain volume measurements: a test-retest dataset
US8208706B2 (en) Functional image presentation
Vavere et al. Diagnostic performance of combined noninvasive coronary angiography and myocardial perfusion imaging using 320 row detector computed tomography: design and implementation of the CORE320 multicenter, multinational diagnostic study
US10074037B2 (en) System and method for determining optimal operating parameters for medical imaging
CN111447874A (zh) 判别装置、抑郁症状的判别方法、抑郁症状的水平的判断方法、抑郁症患者的分类方法、抑郁症状的治疗效果的判断方法和脑活动训练装置
Shinohara et al. Population-wide principal component-based quantification of blood–brain-barrier dynamics in multiple sclerosis
CN1883414A (zh) 用于计划检查的方法和为此的诊断设备
Reith et al. Application of deep learning to predict standardized uptake value ratio and amyloid status on 18F-florbetapir PET using ADNI data
KR102144719B1 (ko) 복수의 학습 모델에 의한 다중 예측 결과를 이용하는 인공지능 기반 파킨슨 병 진단 장치 및 방법
Scott et al. An overview of the quality assurance and quality control of magnetic resonance imaging data for the Ontario Neurodegenerative Disease Research Initiative (ONDRI): pipeline development and neuroinformatics
Malek-Ahmadi et al. Volumetric regional MRI and neuropsychological predictors of motor task variability in cognitively unimpaired, Mild Cognitive Impairment, and probable Alzheimer's disease older adults
Chattopadhyay et al. Comparison of deep learning architectures for predicting amyloid positivity in Alzheimer’s disease, mild cognitive impairment, and healthy aging, from T1-weighted brain structural MRI
WO2022168969A1 (ja) 学習装置、学習済みモデルの生成方法、診断処理装置、コンピュータプログラム及び診断処理方法
Suotunen et al. Visual assessment of [11C] PIB PET in patients with cognitive impairment
Ufuk et al. Comparing three-dimensional zero echo time (3D-ZTE) lung MRI and chest CT in the evaluation of systemic sclerosis-related interstitial lung disease
KR102144720B1 (ko) 기계 학습을 통해 분류된 나이그로좀 1 영역과 복수의 학습 모델에 의한 다중 예측 결과를 이용하는 인공지능 기반 파킨슨 병 진단을 위한 정보제공 방법
KR102708335B1 (ko) 치매 진단을 위한 의료 영상 회귀분석 장치 및 방법
Hassanaly et al. Pseudo-healthy image reconstruction with variational autoencoders for anomaly detection: A benchmark on 3D brain FDG PET
JPWO2022104288A5 (enrdf_load_stackoverflow)
KR102144718B1 (ko) 기계 학습을 통해 분류된 나이그로좀 1 영역을 이용하는 인공지능 기반 파킨슨 병 진단 장치 및 방법
Skattebøl et al. Brain age in multiple sclerosis: a study with deep learning and traditional machine learning
Choi et al. Can patient-specific acquisition protocol improve performance on defect detection task in myocardial perfusion SPECT?

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: 22749840

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2022579638

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22749840

Country of ref document: EP

Kind code of ref document: A1