WO2020262681A1 - 学習装置、方法およびプログラム、医用画像処理装置、方法およびプログラム、並びに判別器 - Google Patents

学習装置、方法およびプログラム、医用画像処理装置、方法およびプログラム、並びに判別器 Download PDF

Info

Publication number
WO2020262681A1
WO2020262681A1 PCT/JP2020/025399 JP2020025399W WO2020262681A1 WO 2020262681 A1 WO2020262681 A1 WO 2020262681A1 JP 2020025399 W JP2020025399 W JP 2020025399W WO 2020262681 A1 WO2020262681 A1 WO 2020262681A1
Authority
WO
WIPO (PCT)
Prior art keywords
learning
teacher label
image
disease
region
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/JP2020/025399
Other languages
English (en)
French (fr)
Japanese (ja)
Inventor
瑞希 武井
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fujifilm Corp
Original Assignee
Fujifilm Corp
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 Fujifilm Corp filed Critical Fujifilm Corp
Priority to JP2021528288A priority Critical patent/JP7170868B2/ja
Publication of WO2020262681A1 publication Critical patent/WO2020262681A1/ja
Priority to US17/553,641 priority patent/US12272049B2/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
    • A61B6/5211—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data
    • A61B6/5217—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data extracting a diagnostic or physiological parameter from medical diagnostic data
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/50—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment specially adapted for specific body parts; specially adapted for specific clinical applications
    • A61B6/501—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment specially adapted for specific body parts; specially adapted for specific clinical applications for diagnosis of the head, e.g. neuroimaging or craniography
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01T—MEASUREMENT OF NUCLEAR OR X-RADIATION
    • G01T1/00—Measuring X-radiation, gamma radiation, corpuscular radiation, or cosmic radiation
    • G01T1/16—Measuring radiation intensity
    • G01T1/161—Applications in the field of nuclear medicine, e.g. in vivo counting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06—COMPUTING OR CALCULATING; 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
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06V10/7747—Organisation of the process, e.g. bagging or boosting
    • 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
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/02—Arrangements for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis
    • A61B6/03—Computed tomography [CT]
    • A61B6/032—Transmission computed tomography [CT]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/02—Arrangements for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis
    • A61B6/03—Computed tomography [CT]
    • A61B6/037—Emission tomography
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10104—Positron emission tomography [PET]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30016—Brain
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images
    • G06V2201/031—Recognition of patterns in medical or anatomical images of internal organs

Definitions

  • the present disclosure is constructed by learning devices, methods and programs for learning a discriminator for extracting a diseased region contained in a medical image, a medical image processing device using the learned discriminator, methods and programs, and learning. It is related to the discriminator.
  • CT Computer Tomography
  • MRI Magnetic Resonance Imaging
  • CAD Computer-Aided Diagnosis
  • a discriminator consisting of a neural network or the like that has been machine-learned by deep learning, etc., and in the brain. It has also been performed to detect diseased areas such as bleeding areas and infarcted areas in.
  • a teacher including a learning image including a disease region and a teacher label in which the disease region is specified by labeling the disease region in the learning image.
  • Data is prepared in advance. Labeling of diseased areas of learning images is done manually by the physician.
  • the learning image is input to the discriminator, the diseased area in the learning image is detected, the difference between the detection result and the teacher label is derived as a loss, and the derived loss is used for the discriminator. Learning takes place.
  • the region to be determined to be the diseased region is often different depending on the labeling doctor. For example, when another doctor sees a diseased area labeled by one doctor, the area wider than the labeled area may be determined as a diseased area. In such a case, if the discriminator is trained using the narrowly labeled disease region as a teacher label, the region that may be detected is learned as the region that should not be detected, so the disease region is overlooked. There is a possibility that it will end up.
  • This disclosure was made in view of the above circumstances, and an object is to prevent oversight of diseased areas.
  • the learning device includes an information acquisition unit that acquires a learning image including a disease area and a first teacher label that identifies a disease area included in the learning image.
  • a teacher label generator that generates at least one second teacher label whose criteria for identifying the disease area are different from the first teacher label. It includes a learning unit that learns a discriminator that detects a diseased region included in a target image based on a learning image, a first teacher label, and at least one second teacher label.
  • the teacher label generation unit may generate at least one second teacher label using the first teacher label.
  • the teacher label generation unit has at least one second teacher based on the distribution of signal values in the region within the first teacher label in the learning image and the position of the first teacher label. It may be one that produces a label.
  • the teacher label generation unit derives a representative value of the signal value of the region in the first teacher label in the learning image, and in the region corresponding to the first teacher label and the learning image.
  • a region in which the signal value of the region adjacent to the region in the first teacher label is within a predetermined range with respect to the representative value may be generated as the second teacher label.
  • the "representative value” for example, an average value, a weighted average value, a median value, a maximum value, a minimum value, and the like can be used.
  • the learning unit detects the learning disease area by inputting the learning image into the discriminator, and the first loss between the learning disease area and the first teacher label, and By deriving the second loss between the learning disease area and the second teacher label, deriving the total loss from the first loss and the second loss, and using the total loss for learning the discriminator, the discriminator It may be something to learn.
  • the learning image may include the brain, and the disease area may be the area of brain disease.
  • the medical image processing device is provided with a disease area detection unit that detects a disease area included in the target medical image by inputting the target medical image to which a discriminator learned by the learning device according to the present disclosure is applied. ..
  • a labeling unit that labels a diseased area detected from a target medical image and a labeling unit are used. It may further include a display control unit that displays the labeled target medical image on the display unit.
  • the discriminator according to the present disclosure is learned by the learning device according to the present disclosure, and detects a disease region included in the target medical image by inputting the target medical image.
  • the learning method obtains a learning image including a disease area and a first teacher label that identifies a disease area included in the learning image. Generate at least one second teacher label whose criteria for identifying the disease area are different from the first teacher label. Based on the learning image, the first teacher label, and at least one second teacher label, a discriminator for detecting a diseased region included in the target image is learned.
  • the medical image processing method detects a disease region included in the target medical image by inputting the target medical image using the discriminator learned by the learning method according to the present disclosure.
  • the learning method and the medical image processing method according to the present disclosure may be provided as a program for executing the computer.
  • Other learning devices include a memory for storing instructions to be executed by a computer and a memory.
  • the processor comprises a processor configured to execute a stored instruction.
  • Based on the learning image, the first teacher label, and at least one second teacher label, a process of learning a discriminator for detecting a diseased region included in the target image is executed.
  • Other medical image processing devices include a memory for storing instructions to be executed by a computer and a memory.
  • the processor comprises a processor configured to execute a stored instruction.
  • a process of detecting a disease region included in the target medical image is executed by inputting the target medical image.
  • Diagram showing brain image and teacher label Diagram showing brain image and teacher label Diagram showing a learning image, a first teacher label and a second teacher label The figure which shows the distribution of CT value in a disease area
  • Diagram for explaining the detection of a diseased area from a target image A flowchart showing the learning process performed in the present embodiment Flowchart showing medical image processing performed in this embodiment Diagram for explaining the detection of a diseased area from a target image
  • FIG. 1 is a hardware configuration diagram showing an outline of a diagnostic support system to which the learning device and the medical image processing device according to the embodiment of the present disclosure are applied.
  • the learning device and the medical image processing device hereinafter referred to as the medical image processing device
  • the three-dimensional image capturing device 2 and the image storage server 3 according to the present embodiment are networked. It is connected in a state where communication is possible via 4.
  • the three-dimensional image capturing device 2 is a device that generates a three-dimensional image representing the site by photographing the site to be diagnosed of the subject, and specifically, a CT device, an MRI device, and a PET (PET). Positron Emission Tomography) equipment, etc.
  • the three-dimensional image generated by the three-dimensional image capturing device 2 is transmitted to the image storage server 3 and stored.
  • the diagnosis target site of the patient as the subject is the brain
  • the three-dimensional imaging device 2 is the CT device
  • a CT image of the head including the brain of the subject is generated as the target image.
  • the learning image used for learning is a CT image of the brain, and the disease area in the learning image is labeled to generate a teacher label.
  • the image storage server 3 is a computer that stores and manages various data, and is equipped with a large-capacity external storage device and database management software.
  • the image storage server 3 communicates with another device via a wired or wireless network 4 to send and receive image data and the like.
  • various data including the image data of the target image generated by the three-dimensional image capturing device 2 are acquired via the network and stored in a recording medium such as a large-capacity external storage device for management.
  • the storage format of the image data and the communication between the devices via the network 4 are based on a protocol such as DICOM (Digital Imaging and Communication in Medicine).
  • DICOM Digital Imaging and Communication in Medicine
  • the medical image processing device 1 is a computer in which the learning program and the medical image processing program of the present embodiment are installed.
  • the computer may be a workstation or personal computer operated directly by the diagnosing doctor, or it may be a server computer connected to them via a network.
  • the learning program and the medical image processing program are stored in the storage device of the server computer connected to the network or the network storage in a state of being accessible from the outside, and are downloaded and installed in the computer upon request. Alternatively, it is recorded and distributed on a recording medium such as a DVD (Digital Versatile Disc) or a CD-ROM (Compact Disc Read Only Memory), and is installed on a computer from the recording medium.
  • a recording medium such as a DVD (Digital Versatile Disc) or a CD-ROM (Compact Disc Read Only Memory)
  • FIG. 2 is a diagram showing a schematic configuration of a medical image processing device realized by installing a learning program and a medical image processing program on a computer.
  • the medical image processing apparatus 1 includes a CPU (Central Processing Unit) 11, a memory 12, and a storage 13 as a standard workstation configuration. Further, a display unit 14 such as a liquid crystal display and an input unit 15 such as a keyboard and a mouse are connected to the medical image processing device 1.
  • a CPU Central Processing Unit
  • memory 12 main memory
  • storage 13 as a standard workstation configuration
  • a display unit 14 such as a liquid crystal display
  • an input unit 15 such as a keyboard and a mouse are connected to the medical image processing device 1.
  • the storage 13 is composed of a hard disk drive or the like, and is a target image to be processed acquired from the image storage server 3 via the network 4, a learning image for learning a neural network as described later, and a teacher for the learning image. Various information including labels and information necessary for processing is stored.
  • the memory 12 stores a learning program and a medical image processing program.
  • the learning program as a process to be executed by the CPU 11, a learning image including a disease area and a first information acquisition process for acquiring a teacher label for specifying a disease area included in the learning image, and a criterion for specifying a disease area are first.
  • a teacher label generation process that generates at least one second teacher label different from the teacher label, and a disease region contained in the target image is detected based on the learning image, the first teacher label, and at least one second teacher label.
  • the learning process for learning the discriminator is specified.
  • the medical image processing program performs a disease area detection process for detecting a disease area included in a target image to be detected of a disease area acquired by an information acquisition process, and labeling the detected disease area.
  • the labeling process to be performed and the display control process for displaying the labeled target image on the display unit 14 are defined.
  • the CPU 11 executes these processes according to the learning program and the medical image processing program, so that the computer can display the information acquisition unit 21, the teacher label generation unit 22, the learning unit 23, the disease area detection unit 24, and the labeling unit 25. It functions as a control unit 26.
  • the information acquisition unit 21 acquires the learning image and the first teacher label that identifies the disease area included in the learning image from the image storage server 3 via an interface (not shown) connected to the network.
  • the target image to be processed is also acquired.
  • the information acquisition unit 21 acquires the learning image, the first teacher label, and the target image from the storage 13. You may do it.
  • the diseased area such as cerebral hemorrhage shows a high or low CT value as compared with the surrounding area.
  • the diseased region 31 shows a higher CT value than the other regions.
  • the teacher label 32 substantially matches the diseased region 31 in the brain image 30 no matter who creates it.
  • the contrast between the diseased area 41 and the surrounding area is often unclear.
  • the contrast is unclear with a broken line.
  • the extent to which the disease area is regarded and labeling differs depending on the doctor performing the labeling. For example, one doctor may give a smaller size teacher label 42, while another doctor may give a larger size teacher label 43.
  • the teacher label generation unit 22 generates at least one second teacher label whose criteria for identifying the disease area are different from the first teacher label. Therefore, the teacher label generation unit 22 derives a representative value of the CT value in the region 55 to which the first teacher label 51 is attached in the acquired learning image 50 as shown in FIG. 5, and derives the representative value of the CT value, and the first teacher label. A region in which the signal value of the region corresponding to 51 and the region adjacent to the region in the first teacher label 51 in the learning image 50 is within a predetermined range with respect to the representative value is generated as the second teacher label. To do.
  • the teacher label generation unit 22 derives the average value ⁇ of the CT values in the region 55 as a representative value, but the present invention is not limited to this. A median value, a weighted average value, a maximum value, a minimum value, or the like may be used as representative values. Further, in the present embodiment, the teacher label generation unit 22 derives the standard deviation ⁇ of the CT value in the region 55 and the position of the center of gravity 56 of the region specified by the first teacher label 51 in the learning image 50.
  • the teacher label generation unit 22 labels a region consisting of pixels having a CT value in the range of ⁇ ⁇ ⁇ among the pixels within a predetermined distance from the position of the center of gravity 56 in the learning image 50, whereby FIG.
  • the second teacher label 52 is generated as shown in.
  • FIG. 6 is a diagram showing the distribution of CT values in the diseased region. As shown in FIG. 6, the CT value of the diseased area is larger than the CT value of the surrounding area, the CT value becomes smaller toward the diseased area, and gradually coincides with the CT value of the area around the diseased area. It is distributed as follows. Therefore, assuming that the first teacher label is given to the range shown by the arrow A shown in FIG.
  • a second teacher label 52 different from the first teacher label 51 is generated as shown in FIG.
  • the CT value in the diseased region is constant, so that the standard deviation ⁇ is substantially 0.
  • the second teacher label 52 generated by the teacher label generation unit 22 is substantially the same as the first teacher label 51.
  • one second teacher label 52 is generated from the first teacher label 51, but a plurality of second teacher labels may be generated.
  • a region consisting of pixels having CT values such as ⁇ ⁇ 0.5 ⁇ , ⁇ ⁇ ⁇ , and ⁇ ⁇ 1.5 ⁇ may be labeled to generate a plurality of second teacher labels 52.
  • the learning unit 23 learns the discriminator 28 that detects the diseased region included in the target image based on the learning image 50, the first teacher label 51, and the second teacher label 52.
  • the discriminator 28 discriminates the diseased region of the brain included in the target image.
  • the discriminator 28 is a convolutional neural network (hereinafter, CNN (Convolutional Neural Network)) which is one of multi-layer neural networks in which a plurality of processing layers are hierarchically connected and deep learning is performed. ) And).
  • CNN Convolutional Neural Network
  • a convolutional neural network consists of a plurality of convolutional layers and a pooling layer.
  • the convolution layer performs convolution processing using various kernels on the input image, and outputs a feature map consisting of feature data obtained by the convolution processing.
  • the convolution layer applies the kernel to the entire input image or the feature map output from the processing layer in the previous stage while shifting the attention pixels of the kernel. Further, the convolutional layer applies an activation function such as a sigmoid function to the convolutional value, and outputs a feature map.
  • the pooling layer reduces the amount of data in the feature map by pooling the feature map output by the convolutional layer, and outputs the feature map with the reduced amount of data.
  • FIG. 7 is a conceptual diagram of learning performed in this embodiment.
  • the learning unit 23 inputs the learning image 50 into the CNN 60 serving as the discriminator 28, and causes the CNN 60 to output the discriminant result 57 of the diseased region in the learning image 50.
  • the discrimination result 57 represents the probability that each pixel of the learning image 50 is a diseased region.
  • the learning unit 23 specifies a region consisting of pixels whose probability is equal to or higher than a predetermined threshold value as a learning disease region 58. Then, the learning unit 23 derives the first loss L1 based on the difference between the first teacher label 51 and the discrimination result 57 of the learning disease region 58.
  • the first loss L1 is the difference between the probability and the above threshold value for the pixel determined not to be the diseased region even though it is the diseased region on the first teacher label 51, and is not the diseased region on the first teacher label 51. However, it is the difference between the above threshold value and the probability for the pixel determined to be the diseased area.
  • the learning unit 23 derives the second loss L2 based on the difference between the second teacher label 52 and the determination result 57.
  • the second loss L2 is the difference between the probability and the above threshold value for the pixel determined not to be the diseased region even though it is the diseased region on the second teacher label 52, and is not the diseased region on the second teacher label 52. However, it is the difference between the above threshold value and the probability for the pixel determined to be the diseased area.
  • the learning unit 23 weights and adds the first loss L1 and the second loss L2 as shown in the following equation (1) to derive the total loss L0 for each pixel of the learning image 50.
  • ⁇ in the equation (1) is a weighting coefficient, and takes a value of 0.5, for example, but is not limited to this.
  • the learning unit 23 uses a large number of learning images 50, the first teacher label 51, and the second teacher label 52 so that the total loss L0 is equal to or less than a predetermined threshold value, and the CNN 60, that is, the discriminator 28 is used.
  • the CNN 60 that is, the discriminator 28 is used.
  • the CNN 60 that is, the discriminator 28 is learned.
  • the discriminator 28 outputs the probability that each pixel of the target image is a diseased region of the brain.
  • the learning unit 23 may perform learning a predetermined number of times instead of learning so that the total loss L0 is equal to or less than a predetermined threshold value.
  • the trained model that outputs the probability that the target image is a disease region included in the target image as a discrimination result is Will be built.
  • the trained model is applied to the disease area detection unit as a discriminator 28.
  • the disease area detection unit 24 detects the disease area included in the target image by using the discriminator 28. That is, the disease region detection unit 24 inputs the target image to the discriminator 28, and causes the discriminator 28 to output the probability that each pixel of the target image is the disease region of the brain. Then, the disease area detection unit 24 detects pixels whose probability exceeds a predetermined threshold value as pixels of the disease area included in the target image.
  • the labeling unit 25 labels the diseased area included in the target image based on the detection result by the diseased area detection unit 24. For example, when the target image 70 including the disease region 71 (indicated by the broken line) is input to the disease region detection unit 24 as shown in FIG. 8, the disease region detection unit 24 uses the disease region 71 included in the target image 70. Is detected.
  • the labeling unit 25 labels the diseased area 71 included in the target image 70 to label the diseased area 71. For example, as shown in FIG. 8, a label 72 is given to the diseased area 71 by changing the color of the diseased area, and labeling is performed. It should be noted that in FIG. 8, changing the color is shown by adding hatching. In addition, labeling may be performed by providing a frame surrounding the diseased area.
  • the display control unit 26 displays the labeled target image on the display unit 14.
  • FIG. 9 is a flowchart showing the learning process performed in the present embodiment. It is assumed that a plurality of learning images and the first teacher label are acquired from the image storage server 3 by the information acquisition unit 21 and stored in the storage 13. First, the information acquisition unit 21 acquires a set of learning images 50 and the first teacher label 51 from the plurality of learning images and the first teacher label stored in the storage 13 (step ST1). Next, the teacher label generation unit 22 generates at least one second teacher label 52 whose criteria for identifying the disease region are different from those of the first teacher label 51 (step ST2).
  • the learning unit 23 inputs the learning image 50, the first teacher label 51, and the second teacher label 52 to the CNN 60 to derive the total loss L0, and the total loss L0 is a predetermined threshold value.
  • the CNN 60 that is, the discriminator 28 is learned as follows (step ST3).
  • step ST1 the process returns to step ST1
  • the next learning image 50 and the first teacher label 51 are acquired from the storage 13, and the processes of step ST2 and step ST3 are repeated.
  • the trained discriminator 28 is constructed.
  • FIG. 10 is a flowchart of medical image processing performed in the present embodiment.
  • the information acquisition unit 21 acquires the target image (step ST11), and the disease area detection unit 24 detects the disease area included in the target image (step ST12).
  • the labeling unit 25 labels the diseased area detected from the target image 70 (step ST13).
  • the display control unit 26 displays the labeled target image on the display unit 14 (step ST14), and ends the process.
  • the learning image 50 including the disease area and the first teacher label 51 for specifying the disease area included in the learning image 50 are acquired, and the criterion for specifying the disease area is the first teacher.
  • the discriminator 28 for detecting the diseased region included in the target image 70 is learned. Therefore, the discriminator 28 detects the diseased region from the target image based not only on the criteria of the first teacher label 51 but also on the criteria of the second teacher label 52.
  • the discriminator 28 since the discriminator 28 is learned using a plurality of teacher labels of different criteria, a disease region in which the contrast with the surroundings is unclear, which is likely to be blurred by the doctor, is defined. , The learned discriminator 28 can detect with a certain tolerance. Therefore, in particular, by generating the second teacher label 52 so as to label a range of diseases larger than the first teacher label 51, a wider range than when learning is performed using only the first teacher label 51.
  • the discriminator 28 can be constructed so that the diseased region of the above can be detected. Therefore, according to the present embodiment, it is possible to prevent the diseased area included in the target image from being overlooked.
  • the probability of indicating that the disease region is output by the discriminator 28 of the disease region detection unit 24 is smaller toward the vicinity of the disease region. Therefore, as shown in FIG. 11, labels having different transparency may be given stepwise according to the probability output by the discriminator 28. In addition, in FIG. 11, it is shown by different hatching that the transparency is different. Moreover, not only the color may be changed stepwise, but also the transparency may be gradually changed. Further, the color may be changed instead of the transparency.
  • the second teacher label 52 generated by the teacher label generation unit 22 is substantially the same as the first teacher label 51. Therefore, in the case of the learning image 50 in which the boundary of the diseased region is clear, the second teacher label 52 may not be generated.
  • the teacher label generation unit 22 determines whether or not the standard deviation ⁇ of the CT value in the diseased area is equal to or less than a predetermined threshold value, and if the standard deviation ⁇ is equal to or less than the threshold value. , The second teacher label 52 may not be generated.
  • the learning of the discriminator 28 is performed using only the first teacher label 51, and the first loss L1 is used as the total loss L0.
  • the second teacher label 52 having a size larger than that of the first teacher label 51 is generated, but the second teacher label 52 having a size smaller than that of the first teacher label 51 is generated. You may.
  • the teacher label generation unit 22 generates the second teacher label 52 from the first teacher label 51, but the present invention is not limited to this.
  • the second teacher label 52 may be generated from the learning image 50.
  • the diseased region of the brain is detected by using the target image as a three-dimensional image including the brain, but the present invention is not limited to this.
  • the techniques of the present disclosure can also be applied to detect diseased areas contained in other structures other than the brain, such as lungs, liver, heart and kidneys.
  • a discriminator 28 capable of detecting the tumor contained in the liver without being overlooked can be constructed.
  • a discriminator 28 capable of detecting the lung nodule contained in the lung without being overlooked can be constructed.
  • a three-dimensional medical image is used as the target image, but the present invention is not limited to this.
  • the individual tomographic images that make up the three-dimensional medical image may be used as the target image.
  • a two-dimensional X-ray image acquired by simple X-ray photography may be used as the target image.
  • a learning image and a first teacher label corresponding to the type of the target image are prepared, and the CNN 60, that is, the discriminator 28 is trained.
  • CNN60 is used as the discriminator 28, but the present invention is not limited to this. If it is a neural network composed of a plurality of processing layers, a deep neural network (DNN (Deep Neural Network)), a recurrent neural network (RNN (Recurrent Neural Network)), and the like can be used.
  • DNN Deep Neural Network
  • RNN Recurrent Neural Network
  • a processing unit that executes various processes such as an information acquisition unit 21, a teacher label generation unit 22, a learning unit 23, a disease area detection unit 24, a labeling unit 25, and a display control unit 26.
  • various processors Processors
  • the various processors include a CPU, which is a general-purpose processor that executes software (program) and functions as various processing units, and a circuit after manufacturing an FPGA (Field Programmable Gate Array) or the like.
  • Dedicated electricity which is a processor with a circuit configuration specially designed to execute specific processing such as programmable logic device (PLD), ASIC (Application Specific Integrated Circuit), which is a processor whose configuration can be changed. Circuits and the like are included.
  • One processing unit may be composed of one of these various processors, or a combination of two or more processors of the same type or different types (for example, a combination of a plurality of FPGAs or a combination of a CPU and an FPGA). ) May be configured. Further, a plurality of processing units may be configured by one processor.
  • one processor is configured by combining one or more CPUs and software. There is a form in which this processor functions as a plurality of processing units.
  • SoC System On Chip
  • the various processing units are configured by using one or more of the various processors as a hardware structure.
  • circuitry in which circuit elements such as semiconductor elements are combined can be used.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • General Physics & Mathematics (AREA)
  • Public Health (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Radiology & Medical Imaging (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Primary Health Care (AREA)
  • Epidemiology (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Optics & Photonics (AREA)
  • Quality & Reliability (AREA)
  • Multimedia (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Surgery (AREA)
  • Veterinary Medicine (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Neurosurgery (AREA)
  • Dentistry (AREA)
  • Neurology (AREA)
  • Spectroscopy & Molecular Physics (AREA)
PCT/JP2020/025399 2019-06-28 2020-06-26 学習装置、方法およびプログラム、医用画像処理装置、方法およびプログラム、並びに判別器 Ceased WO2020262681A1 (ja)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2021528288A JP7170868B2 (ja) 2019-06-28 2020-06-26 学習装置、方法およびプログラム、医用画像処理装置、方法およびプログラム、並びに判別器
US17/553,641 US12272049B2 (en) 2019-06-28 2021-12-16 Learning device, method, and program, medical image processing apparatus, method, and program, and discriminator

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2019121015 2019-06-28
JP2019-121015 2019-06-28

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/553,641 Continuation US12272049B2 (en) 2019-06-28 2021-12-16 Learning device, method, and program, medical image processing apparatus, method, and program, and discriminator

Publications (1)

Publication Number Publication Date
WO2020262681A1 true WO2020262681A1 (ja) 2020-12-30

Family

ID=74060193

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2020/025399 Ceased WO2020262681A1 (ja) 2019-06-28 2020-06-26 学習装置、方法およびプログラム、医用画像処理装置、方法およびプログラム、並びに判別器

Country Status (3)

Country Link
US (1) US12272049B2 (https=)
JP (1) JP7170868B2 (https=)
WO (1) WO2020262681A1 (https=)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2025531808A (ja) * 2023-07-24 2025-09-25 オンコソフト インコーポレイテッド 医療映像分割方法及びこれを遂行する装置
JP2025534224A (ja) * 2023-08-29 2025-10-15 オンコソフト インコーポレイテッド 医療映像分割方法及びこれを遂行する装置

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2018061771A (ja) * 2016-10-14 2018-04-19 株式会社日立製作所 画像処理装置、及び画像処理方法
JP2019509813A (ja) * 2016-03-16 2019-04-11 ハートフロー， インコーポレイテッド 冠動脈において健全な管腔径を推定し狭窄を定量化するためのシステム及び方法
US20190192096A1 (en) * 2017-12-21 2019-06-27 Beijing Curacloud Technology Co., Ltd. Method and device for generating anatomical labels for a physiological tree structure

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019146358A1 (ja) * 2018-01-24 2019-08-01 富士フイルム株式会社 学習システム、方法及びプログラム
US11521742B2 (en) * 2018-07-18 2022-12-06 SCA Robotics Methods of implementing an artificial intelligence based neuroradiology platform for neurological tumor identification and for T-Cell therapy initiation and tracking and related precision medical treatment predictive modeling
JP6996633B2 (ja) * 2018-08-06 2022-01-17 株式会社島津製作所 教師ラベル画像修正方法、学習済みモデルの作成方法および画像解析装置
JPWO2020174862A1 (ja) * 2019-02-28 2021-12-23 ソニーグループ株式会社 情報処理装置、情報処理方法および情報処理システム
WO2020194662A1 (ja) * 2019-03-28 2020-10-01 オリンパス株式会社 情報処理システム、内視鏡システム、学習済みモデル、情報記憶媒体、情報処理方法及び学習済みモデルの製造方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2019509813A (ja) * 2016-03-16 2019-04-11 ハートフロー， インコーポレイテッド 冠動脈において健全な管腔径を推定し狭窄を定量化するためのシステム及び方法
JP2018061771A (ja) * 2016-10-14 2018-04-19 株式会社日立製作所 画像処理装置、及び画像処理方法
US20190192096A1 (en) * 2017-12-21 2019-06-27 Beijing Curacloud Technology Co., Ltd. Method and device for generating anatomical labels for a physiological tree structure

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
HAN, JIANGFAN ET AL.: "Deep Self-Learning From Noisy Labels", DEEP SELF-LEARNING FROM NOISY LABELS, 6 August 2019 (2019-08-06), XP033723138 *
NATARAJAN, NAGARAJAN ET AL.: "Leaning with Noisy Labels", NIPS'13: PROCEEDINGS OF THE 26TH INTERNATIONAL CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEMS, vol. 1, 2013, pages 1196 - 1204, XP055603359 *
SHIMABARA YUKI: "Development of cerebral aneurysm detection software utilizing deep learning", MEDICAL IMAGING AND INFORMATION SCIENCES, vol. 34, no. 2, 2017, pages 103 - 104 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2025531808A (ja) * 2023-07-24 2025-09-25 オンコソフト インコーポレイテッド 医療映像分割方法及びこれを遂行する装置
JP2025534224A (ja) * 2023-08-29 2025-10-15 オンコソフト インコーポレイテッド 医療映像分割方法及びこれを遂行する装置

Also Published As

Publication number Publication date
US20220108451A1 (en) 2022-04-07
JP7170868B2 (ja) 2022-11-14
JPWO2020262681A1 (https=) 2020-12-30
US12272049B2 (en) 2025-04-08

Similar Documents

Publication Publication Date Title
JP7018856B2 (ja) 医用画像処理装置、方法およびプログラム
US11182901B2 (en) Change detection in medical images
JP7129869B2 (ja) 疾患領域抽出装置、方法及びプログラム
US12198343B2 (en) Multi-modal computer-aided diagnosis systems and methods for prostate cancer
US11244455B2 (en) Apparatus, method, and program for training discriminator discriminating disease region, discriminator discriminating disease region, disease region discrimination apparatus, and disease region discrimination program
JP7339270B2 (ja) 医用画像処理装置、方法およびプログラム
US11049251B2 (en) Apparatus, method, and program for learning discriminator discriminating infarction region, discriminator for discriminating infarction region, and apparatus, method, and program for discriminating infarction region
JP7342120B2 (ja) 学習装置、方法およびプログラム、クラス分類装置、方法およびプログラム、並びに学習済みモデル
US11176413B2 (en) Apparatus, method, and program for training discriminator discriminating disease region, discriminator discriminating disease region, disease region discrimination apparatus, and disease region discrimination program
JP7007469B2 (ja) 医療文書作成支援装置、方法およびプログラム、学習済みモデル、並びに学習装置、方法およびプログラム
US12272049B2 (en) Learning device, method, and program, medical image processing apparatus, method, and program, and discriminator
WO2020241857A1 (ja) 医療文書作成装置、方法およびプログラム、学習装置、方法およびプログラム、並びに学習済みモデル
JP7109345B2 (ja) 優先度判定装置、方法およびプログラム
WO2020110520A1 (ja) 類似度決定装置、方法およびプログラム
EP3965117A1 (en) Multi-modal computer-aided diagnosis systems and methods for prostate cancer
WO2019216125A1 (ja) 梗塞領域を判別する判別器の学習装置、方法およびプログラム、梗塞領域を判別する判別器、並びに梗塞領域判別装置、方法およびプログラム
JP7479546B2 (ja) 表示装置、方法およびプログラム
EP4657379A1 (en) Hybrid intelligence for knowledge driven synthesis of realistic lesions
WO2022137855A1 (ja) 情報処理装置、方法およびプログラム
WO2021060461A1 (ja) 画像処理装置、方法およびプログラム
WO2022054541A1 (ja) 画像処理装置、方法およびプログラム

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: 20831468

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021528288

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: 20831468

Country of ref document: EP

Kind code of ref document: A1