Program, information processing method, information processing device, and model generation method

Abstract

Provided is a program capable of acquiring increase information related to the presence or absence of an increase in cerebral aneurysm, on the basis of a Magnetic Resonance Angiography (MRA) image of a brain. A computer acquires MRA images of the brain. If MRA images of a brain have been input, the computer inputs the acquired MRA images of the brain to a learning model that outputs increase information related to the presence or absence of an increase in cerebral aneurysm, and the computer thereby acquires increase information related to the presence or absence of an increase in a cerebral aneurysm in the input MRA images.

Description

Program, information processing method, information processing device, and model generation method

[0001] The present invention relates to a program, an information processing method, an information processing device, and a model generation method.

[0002] In recent years, development of a technology for detecting diseases in a target region of a patient based on medical images of the target region has been progressing. For example, Patent Literature 1 discloses a technology for detecting lesions in tubular tissues such as blood vessels or the digestive tract from MRI images taken using an MRI (Magnetic Resonance Imaging) device. The technology disclosed in Patent Literature 1 can detect, for example, cerebral aneurysms from MRI images of the head.

[0003] JP 2024-18421 A

[0004] Because rupture of a cerebral aneurysm can cause subarachnoid hemorrhage, the risk of rupture is predicted, and a decision is made whether to perform preventative surgery or follow-up observation based on the rupture risk. For example, the Unruptured Cerebral Aneurysm Study of Japan (UCAS-Japan) score, calculated from the patient's age and gender, the size, location, and shape of the cerebral aneurysm, is used to predict the risk of rupture. However, it is difficult to prevent excessive surgical treatment or rupture during follow-up observation. To prevent these, it is considered important to predict whether or not a cerebral aneurysm will grow and rupture. The technology disclosed in Patent Document 1 can detect the presence or absence of a cerebral aneurysm, but has the problem of being unable to predict the growth of the cerebral aneurysm. Furthermore, the UCAS-Japan score is a score for predicting the risk of rupture, but it does not predict the growth of the cerebral aneurysm.

[0005] In one aspect, an object of the present invention is to provide a program or the like that can acquire growth information regarding the presence or absence of growth of a cerebral aneurysm based on MRA (Magnetic Resonance Angiography) images of the brain.

[0006] A program according to one aspect acquires an MRA image of the brain, and causes a computer to execute a process of acquiring enlargement information in the input MRA image by inputting the acquired MRA image of the brain into a learning model that outputs enlargement information regarding whether or not cerebral aneurysms have enlarged in the input MRA image when the MRA image of the brain is input.

[0007] In one aspect, growth information regarding the presence or absence of growth of a cerebral aneurysm can be obtained based on an MRA image of the brain.

[0008] 1 is a block diagram showing an example of the configuration of an information processing device. FIG. 1 is an explanatory diagram showing an example of the configuration of a first growth estimation model. FIG. 2 is an explanatory diagram showing an example of the configuration of the first growth estimation model. A flowchart showing an example of a generation processing procedure for the first growth estimation model. A flowchart showing an example of a processing procedure for estimating cerebral aneurysm growth. An explanatory diagram showing an example of a screen. FIG. 2 is an explanatory diagram showing an example of the configuration of a second growth estimation model. A flowchart showing an example of a processing procedure for estimating cerebral aneurysm growth of embodiment 2. An explanatory diagram showing an example of the configuration of a third growth estimation model. An explanatory diagram showing an example of the configuration of a fourth growth estimation model. A flowchart showing an example of a processing procedure for estimating cerebral aneurysm growth of embodiment 3. An explanatory diagram showing an example of the configuration of a cerebral aneurysm detection model. A flowchart showing an example of a processing procedure for estimating cerebral aneurysm growth of embodiment 4. An explanatory diagram showing an example of a screen. A flowchart showing another example of the processing procedure for estimating cerebral aneurysm growth of embodiment 4. An explanatory diagram showing an example of a screen. An explanatory diagram showing an example of a screen. An explanatory diagram showing an explanatory diagram showing an effect obtained by using the information processing devices of embodiments 1 to 4. An explanatory diagram showing an effect obtained by using the information processing devices of embodiments 1 to 4.

[0009] Hereinafter, a program, an information processing method, an information processing device, and a model generation method according to the present disclosure will be described in detail with reference to the drawings illustrating embodiments thereof.

[0010] (Embodiment 1) In this embodiment, an information processing device is described that uses artificial intelligence (AI) to estimate whether a cerebral aneurysm (unruptured cerebral aneurysm) occurring in a subject's cerebral blood vessels will grow in the future, based on magnetic resonance angiography (MRA) images of the cerebral blood vessels obtained by photographing the subject's head (brain) with an MRI device. The brain MRA image is obtained by extracting the cerebral blood vessels from an image taken with an MRI device and creating a three-dimensional image. The brain MRA image is used to examine the presence or absence of an unruptured cerebral aneurysm that poses a risk of subarachnoid hemorrhage. Note that growth of a cerebral aneurysm refers to, for example, an increase in the size (e.g., maximum diameter) of the cerebral aneurysm of 1 mm or more. However, the amount of increase is not limited to 1 mm and may be an increase of more than a predetermined value set by a doctor or the like.

[0011] FIG. 1 is a block diagram showing an example configuration of an information processing device. The information processing device 10 is a device capable of various information processing and information transmission / reception, such as a personal computer, server computer, workstation, or tablet PC (personal computer). The information processing device 10 is installed and used in a medical institution, testing institution, research institution, or the like. The information processing device 10 is not limited to a single computer, but may also be a computer system consisting of multiple computers and peripheral devices, a multi-computer consisting of multiple computers, or a virtual machine virtually constructed by software within a single device. When the information processing device 10 is configured as a server computer, the information processing device 10 may be a local server installed in a medical institution or the like, or a cloud server connected to a network such as the Internet. The information processing device 10 may also be an information processing device such as a smartphone, a mobile phone, a wearable device such as an Apple Watch (registered trademark), or a tablet terminal. In this embodiment, the information processing device 10 is described as a single computer.

[0012] Because unruptured cerebral aneurysms can cause subarachnoid hemorrhage if they rupture, a decision is made as to whether to perform preventive surgery or follow-up observation based on the risk of rupture. However, it is not easy to reliably prevent excessive surgical treatment or rupture during follow-up observation. Therefore, the information processing device 10 of this embodiment acquires information on the growth of cerebral aneurysms, which serves as the basis for deciding whether to perform preventive surgery or follow-up observation, based on MRA images of the brain.

[0013] The information processing device 10 includes a control unit 11, a storage unit 12, a communication unit 13, an input unit 14, a display unit 15, a reading unit 16, etc., and these units are connected via a bus. The control unit 11 includes one or more processors (arithmetic processing devices), such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), a GPU (Graphics Processing Unit), an AI chip (AI semiconductor), etc. The control unit 11 executes the processing to be performed by the information processing device 10 by appropriately reading and executing a program P stored in the storage unit 12. Note that when the control unit 11 includes multiple processors, each processing may be executed by the same processor, or each processing may be executed by a different processor.

[0014] The storage unit 12 includes RAM (Random Access Memory), flash memory, HDD (Hard Disk Drive), SSD (Solid State Drive), etc. The storage unit 12 stores a program P (program product, computer program) executed by the control unit 11 and various data required for executing the program P. The storage unit 12 also temporarily stores data generated when the control unit 11 executes the program P. The storage unit 12 also stores a first growth estimation model M1 that has learned training data through machine learning. The first growth estimation model M1 (learning model) is a model that, when an MRA image of the brain is input, is trained to output information indicating whether a cerebral aneurysm (unruptured cerebral aneurysm) contained in the MRA image will grow in the future (growth information regarding the presence or absence of growth of the cerebral aneurysm). The first growth estimation model M1 is expected to be used as a program module constituting artificial intelligence software. The first growth estimation model M1 performs a predetermined calculation on an input value and outputs the calculation result, and the storage unit 12 stores information defining the first growth estimation model M1, such as information on the layers included in the first growth estimation model M1, information on the nodes constituting each layer, and weights (coupling coefficients) between nodes. The storage unit 12 may be composed of multiple storage devices, and part of the storage unit 12 may be another storage device connected to the information processing device 10, or another storage device with which the information processing device 10 can communicate.

[0015] The communication unit 13 is a communication module for performing processes related to wired or wireless communication, and transmits and receives information to and from other devices via a network N. The network N may be a dedicated network, the Internet, a public communication line, or a local area network (LAN) established in a medical institution or other facility where the information processing device 10 is installed. The input unit 14 accepts operation inputs from a user of the information processing device 10 and sends control signals corresponding to the operation content to the control unit 11. The display unit 15 is a liquid crystal display, an organic EL display, or the like, and displays various information in accordance with instructions from the control unit 11. A portion of the input unit 14 and the display unit 15 may be an integrated touch panel. Note that the input unit 14 and the display unit 15 are not essential; the information processing device 10 may be configured to accept operations via a connected computer or to output information to be displayed to an external display device.

[0016] The reading unit 16 reads information stored in a portable storage medium 10a, such as a CD (Compact Disc), a DVD (Digital Versatile Disc), a USB (Universal Serial Bus) memory, an SD card, or a CompactFlash (registered trademark). The program P (program product) and various data stored in the storage unit 12 may be read by the control unit 11 from the portable storage medium 10a via the reading unit 16 and stored in the storage unit 12. The program P and various data may be written to the storage unit 12 during the manufacturing stage of the information processing device 10, or may be downloaded by the control unit 11 from another device via the communication unit 13 and stored in the storage unit 12.

[0017] In this embodiment, the program P may be located on a single computer, or at one site, or may be deployed to run on multiple computers distributed across multiple sites and interconnected via a network N.

[0018] 2A and 2B are explanatory diagrams showing an example of the configuration of the first growth estimation model M1. FIG. 2A shows the first growth estimation model M1 of this embodiment, and FIG. 2B shows a modified version of the first growth estimation model M1 (first growth estimation model M1a). The first growth estimation model M1 in FIG. 2A is trained to use an MRA image of a subject's brain as input data, perform a calculation based on the input data to determine (estimate) whether a cerebral aneurysm in the MRA image will grow in the future, and output the calculation result. The MRA image input to the first growth estimation model M1 is a cerebral aneurysm region cut out (extracted) from an MRA image taken using an MRI device, including the region of the cerebral aneurysm. The cerebral aneurysm region can be, for example, an MRA image of 64 pixels x 64 pixels x 32 pixels, with 64 pixels in the left-right direction, 64 pixels in the up-down direction, and 32 pixels in the front-to-back direction of the subject, but the cerebral aneurysm region input to the first growth estimation model M1 is not limited to this size. Hereinafter, the MRA image of the cerebral aneurysm region input to the first augmented estimation model M1 may be simply referred to as an MRA image.

[0019] The first augmentation estimation model M1 is configured using a CNN (Convolutional Neural Network)-based model such as ResNet (Residual Network), VGG, Efficient Net, etc. Note that the first augmentation estimation model M1 is not limited to a CNN-based model, and may be configured using other algorithms such as Transformer, or may be configured by combining multiple algorithms.

[0020] The first growth estimation model M1 has an input layer, an intermediate layer, and an output layer. The input layer has multiple input nodes, and pixel values ​​of each pixel included in the cerebral aneurysm region extracted from an MRA image of the subject's brain are input via each input node. The intermediate layer uses various functions, thresholds, etc. to extract image features from the MRA image of the cerebral aneurysm region input via the input layer, calculates output values ​​from the extracted image features, and outputs the calculated output values ​​to the output layer. The output layer has two output nodes, each associated with a predetermined discrimination result, and outputs a probability (confidence) indicating that the corresponding discrimination result should be determined. In the example of FIG. 2A , one output node outputs the probability that the cerebral aneurysm in the input cerebral aneurysm region should be determined to grow in the future (information indicating growth), and the other output node outputs the probability that the cerebral aneurysm in the input cerebral aneurysm region should not be determined to grow in the future (information indicating no growth). The output value of each output node is, for example, a value between 0 and 1.0, and the sum of the confidence levels output from each output node is 1.0. With the above-described configuration, when an MRA image (cerebral aneurysm region) is input, the first growth estimation model M1 outputs information (growth information) indicating whether the cerebral aneurysm in the input MRA image will grow in the future.

[0021] When the information processing device 10 uses the first growth estimation model M1 shown in FIG. 2A , it identifies the output node that outputs the largest output value (certainty factor) among the output values ​​from each output node, and identifies the discrimination result associated with the identified output node as the discrimination result to be estimated. Note that the first growth estimation model M1 may be configured to have a single output node that outputs a discrimination result with a high confidence factor (information indicating whether or not cerebral aneurysm growth is occurring) instead of two output nodes that output confidence factors for each discrimination result. Furthermore, the first growth estimation model M1 may be configured to have a single output node that outputs the possibility of cerebral aneurysm growth in the input MRA image, as in the first growth estimation model M1a shown in FIG. 2B .

[0022] The first growth estimation model M1 is generated by machine learning using training data that associates training MRA images (cerebral aneurysm regions) with correct labels (growth information, e.g., a label indicating growth is 1 and a label indicating no growth is 0) indicating whether the cerebral aneurysm in the MRA image will grow in the future. The training data is generated, for example, for a person in whom a cerebral aneurysm was discovered and whose cerebral aneurysm grew during several years to 10 years of follow-up (a follow-up period), by assigning a label indicating that the cerebral aneurysm grew (e.g., 1) to a cerebral aneurysm region extracted from an MRA image of the cerebral aneurysm before the growth. The training data is also generated, for example, for a person in whom a cerebral aneurysm was discovered and whose cerebral aneurysm did not grow during several years to 10 years of follow-up, by assigning a label indicating that the cerebral aneurysm did not grow (e.g., 0) to a cerebral aneurysm region extracted from an MRA image of the cerebral aneurysm. The training data may include training data in which a label indicating that the cerebral aneurysm did not grow is added to MRA images of people without cerebral aneurysms. The training data generated in this manner is stored in a training DB (not shown) prepared in the storage unit 12, for example, and is used during the learning process.

[0023] The first growth estimation model M1 learns to predict whether the output value from the output node corresponding to the discrimination result indicated by the correct label approaches 1 and the output value from the other output node approaches 0 when an MRA image (cerebral aneurysm region) included in the training data is input. During the learning process, the first growth estimation model M1 performs calculations based on the input MRA image to calculate output values ​​from each output node. The first growth estimation model M1 then compares the calculated output values ​​of each output node with values ​​corresponding to the correct label (1 for the output node corresponding to the discrimination result indicated by the correct label, and 0 for the other output node) and optimizes parameters used in the calculation process so that the two values ​​approximate each other. For example, the first growth estimation model M1 optimizes parameters such as the weights (coupling coefficients) between nodes and the coefficients of functions used in each node using backpropagation, steepest descent, or the like. This results in a first growth estimation model M1 that, when an MRA image including a cerebral aneurysm region is input, estimates whether the cerebral aneurysm in the MRA image will grow in the future and outputs the estimation result.

[0024] The first growth estimation model M1 is not limited to the configuration shown in FIG. 2A . For example, the MRA image input to the first growth estimation model M1 is not limited to a 64-pixel x 64-pixel x 32-pixel cerebral aneurysm region, but may instead be the entire MRA image captured using an MRI device. Furthermore, the information output from the first growth estimation model M1 is not limited to the confidence level regarding whether the cerebral aneurysm in the input MRA image will grow in the future, but may also output the possibility (probability) of cerebral aneurysm growth, the risk level of cerebral aneurysm growth, etc. Even when using such a first growth estimation model M1, the information processing device 10 can estimate whether the cerebral aneurysm in the input MRA image will grow in the future based on the output data from the first growth estimation model M1.

[0025] The first growth estimation model M1 may be trained by the information processing device 10 or by another learning device. The trained first growth estimation model M1 generated by training on the other learning device is downloaded from the learning device to the information processing device 10 via the network N or the portable storage medium 10a, for example, and stored in the storage unit 12.

[0026] The process of generating the first growth estimation model M1 by learning the training data will be described below. Fig. 3 is a flowchart showing an example of the process of generating the first growth estimation model M1. The following process is performed by the control unit 11 of the information processing device 10 in accordance with the program P stored in the storage unit 12, but may also be performed by another learning device.

[0027] The control unit 11 of the information processing device 10 first acquires each piece of information to be used in the training data (specifically, MRA images of the brain and labels indicating whether or not cerebral aneurysms in the MRA images have grown during the follow-up period) to generate training data, and then uses the generated training data to train the first growth estimation model M1. The labels for the MRA images use values ​​determined by experts such as doctors (e.g., 1 if growth has occurred, 0 if no growth has occurred).

[0028] The control unit 11 of the information processing device 10 acquires an MRA image of the brain of a subject (patient, subject) and a label assigned to the MRA image (S11). The MRA image and label can be acquired from an image server or electronic medical record server that stores medical images captured by a medical imaging device, such as an MRI device. The label for each MRA image may be assigned to the MRA image and stored in the image server or electronic medical record server, or may be assigned after the MRA image is acquired from the image server or electronic medical record server. The MRA image and label may be acquired in advance from the image server or electronic medical record server and stored in the image DB of the storage unit 12. In this case, the control unit 11 simply reads the MRA image and label from the storage unit 12 (image DB). The control unit 11 is not limited to acquiring MRA images from an image server or electronic medical record server; it may also acquire the MRA image directly from the MRI device or via the network N.

[0029] The control unit 11 extracts a cerebral aneurysm region containing a cerebral aneurysm from the acquired MRA image (S12). Here, the cerebral aneurysm region has already been detected in the MRA image, and the center of gravity (center) of the detected cerebral aneurysm region is used as the center of gravity to extract a cerebral aneurysm region, for example, 64 pixels x 64 pixels x 32 pixels in the left-right x up-down x front-back directions of the subject. The control unit 11 then associates the label acquired in step S11 with the extracted cerebral aneurysm region as a correct label (augmented information), and stores the associated data in the memory unit 12 as training data (S13). The control unit 11 stores the training data in a training database provided in the memory unit 12, for example.

[0030] The control unit 11 determines whether there are any unprocessed MRA images that have not been subjected to the above-described processing among the MRA images used to generate the training data (S14). If it is determined that there are any unprocessed MRA images (S14: YES), the control unit 11 returns to step S11, performs the processing of steps S11 to S13 on the unprocessed MRA images, and stores training data in the storage unit 12 that associates the cerebral aneurysm regions and correct labels extracted from the MRA images of other subjects. The control unit 11 repeats the processing of steps S11 to S14 until it determines that there are no unprocessed MRA images. As a result, training data used to learn the first augmented estimation model M1 is generated and stored in the training DB based on the MRA images prepared for generating the training data and the labels assigned to each MRA image.

[0031] If the control unit 11 determines that there are no unprocessed MRA images (S14: NO), it performs training of the first augmented estimation model M1 using the training data stored in the training DB as described above. In the training data generation process described above, the control unit 11 may increase the number of MRA images by performing data expansion on the MRA images prepared for generating the training data through image processing such as rotation, translation, enlargement or reduction, and inversion, and then perform the processes of steps S11 to S14 described above on the MRA images added by data expansion. In this case, it is possible to create a sufficient amount of training data.

[0032] In the learning process of the first growth estimation model M1, the control unit 11 reads one of the training data stored in the training DB (S15) and performs the learning process of the first growth estimation model M1 based on the read training data (S16). Here, the control unit 11 inputs an MRA image (cerebral aneurysm region) included in the training data into the first growth estimation model M1 and obtains an output value from each output node. The control unit 11 compares the output value of each output node with a value corresponding to the correct label (1 for the output node corresponding to the correct label, 0 for the other output nodes), and optimizes parameters such as the weights between nodes in the first growth estimation model M1 using, for example, backpropagation algorithm, so that the two values ​​approximate each other.

[0033] The control unit 11 determines whether or not there is unprocessed training data that has not yet undergone learning processing among the training data stored in the training DB (S17). If it is determined that there is unprocessed training data (S17: YES), the control unit 11 returns to step S15 and performs the processing of steps S15 to S16 on the unprocessed training data. The control unit 11 repeats the processing of steps S15 to S17 until it determines that there is no unprocessed training data. As a result, the learning processing of the first augmented estimation model M1 is performed using the training data stored in the training DB. If it is determined that there is no unprocessed training data (S17: NO), the control unit 11 ends the series of processing.

[0034] The above-described learning process generates a first growth estimation model M1 that, when an MRA image of a brain containing a cerebral aneurysm (a cerebral aneurysm region in the MRA image) is input, outputs information indicating whether the cerebral aneurysm will grow in the future (growth information regarding the presence or absence of cerebral aneurysm growth). Using this first growth estimation model M1, the information processing device 10 can estimate whether a cerebral aneurysm in an MRA image of a subject's (patient's) brain will grow in the future. In the above-described process, the training data generation process in steps S11 to S14 and the first growth estimation model M1 generation process in steps S15 to S17 may be performed by separate devices. The first growth estimation model M1 can be further optimized by repeatedly performing the learning process using the training data described above. Furthermore, the already-trained first growth estimation model M1 may be fine-tuned by the above-described learning process using training data for each medical institution or MRI device. In this case, a first growth estimation model M1 tailored to each medical institution or MRI device can be generated.

[0035] Next, a process for estimating whether a cerebral aneurysm will grow in the future based on an MRA image of a subject with a cerebral aneurysm will be described using the first growth estimation model M1 generated by the above-described process. Fig. 4 is a flowchart showing an example of the process procedure for estimating the growth of a cerebral aneurysm, and Fig. 5 is an explanatory diagram showing an example screen.

[0036] The control unit 11 of the information processing device 10 acquires an MRA image of the brain of a subject (examinee), such as a patient (S21). The control unit 11 acquires the MRA image of the subject (subject) for whom cerebral aneurysm growth is to be estimated, for example, from an electronic medical record server. The control unit 11 then extracts the cerebral aneurysm region from the acquired MRA image (S22). Step S22 is the same process as step S12 in FIG. 3. Here, the cerebral aneurysm region has already been detected in the MRA image, and the cerebral aneurysm region is extracted based on the center of gravity (center) of the detected cerebral aneurysm region. Based on the extracted cerebral aneurysm region, the control unit 11 determines whether the cerebral aneurysm will grow in the future (presence or absence of cerebral aneurysm growth) (S23). Specifically, the control unit 11 inputs the cerebral aneurysm region into a first growth estimation model M1 and determines whether the cerebral aneurysm will grow based on the output value from the first growth estimation model M1. For example, the control unit 11 identifies the output node that output the largest output value (certainty level) among the output values ​​from the first growth estimation model M1, and identifies the discrimination result associated with the identified output node as the judgment result of whether or not the cerebral aneurysm has grown.

[0037] The control unit 11 stores the determination result in the storage unit 12 or the electronic medical record data in the electronic medical record server (S24). The control unit 11 then generates a screen displaying the determination result, outputs it to, for example, the display unit 15 (S25), and displays it on the display unit 15, thereby completing the process. For example, the control unit 11 generates and displays a determination result screen as shown in FIG. 5. The screen shown in FIG. 5 displays the subject's identification information (e.g., patient ID, patient name, etc.), MRA images and their capture dates and times, and an estimation result regarding the presence or absence of cerebral aneurysm enlargement based on the MRA images. In the example of FIG. 5, "High probability of cerebral aneurysm enlargement" is displayed. However, for example, the confidence level of cerebral aneurysm enlargement output from the first enlargement estimation model M1 may be displayed as the possibility of cerebral aneurysm enlargement. Furthermore, for example, if comments to be presented to a doctor or the like are stored in the storage unit 12 in association with the MRA images or the estimation results, the control unit 11 may read the comments corresponding to the estimation results from the storage unit 12 and display them on the determination result screen.

[0038] The above-described process allows, when a cerebral aneurysm is discovered based on MRA images, an estimate of whether the cerebral aneurysm will grow in the future and present this estimate to a physician. Therefore, physicians can appropriately predict whether the aneurysm will grow, which is a precursor to rupture, based on the presented estimate, in addition to the size and location of the aneurysm. As a result, if the aneurysm is likely to grow, early intervention with preventive treatment, such as surgery to prevent rupture, can prevent the aneurysm from rupturing. If the aneurysm is unlikely to grow, the follow-up period (the period of observation based on MRA images) can be shortened, thereby preventing excessive treatment and rupture during observation.

[0039] In this embodiment, the training data generation process, the learning process of the first growth estimation model M1, and the cerebral aneurysm growth estimation process using the first growth estimation model M1 are not limited to being performed locally by the information processing device 10. For example, an information processing device may be provided to perform each of the above processes. Alternatively, a server may be provided to perform the training data generation process and the learning process of the first growth estimation model M1. In this case, the information processing device 10 may be configured to transmit MRA images and labels used for the training data to the server, and the server may generate training data from the MRA images and labels, generate the first growth estimation model M1 through a learning process using the generated training data, and transmit the model to the information processing device 10. Thus, the information processing device 10 can perform the process of estimating the presence or absence of cerebral aneurysm growth using the first growth estimation model M1 obtained from the server. Alternatively, a server may be provided to perform the process of estimating the presence or absence of cerebral aneurysm growth using the first growth estimation model M1. In this case, the information processing device 10 can be configured to transmit MRA images of the subject to a server, and the server can perform processing to estimate the presence or absence of cerebral aneurysm growth using the first growth estimation model M1, and transmit the estimation result to the information processing device 10. Even with such a configuration, processing similar to that of the present embodiment described above is possible, and similar effects can be obtained.

[0040] In the above-described embodiment, an example has been described in which a configuration is used to estimate whether a cerebral aneurysm will grow based on a brain MRA image. However, this is not limiting. For example, the first growth estimation model M1 may be configured to input a brain MRA image and estimate whether each symptom of a brain disease, such as a brain tumor, cerebral vascular malformation, or stenosis, will grow in the future in the MRA image. Furthermore, the first growth estimation model M1 may be configured to input a chest MRA image and estimate whether an aortic aneurysm or coronary aneurysm will grow in the future in the MRA image. In this case, by using the first growth estimation model M1, it is possible to obtain an estimation result of whether an aortic aneurysm or coronary aneurysm will grow from the chest MRA image. Furthermore, the first growth estimation model M1 may be configured to input an abdominal MRA image and estimate whether an abdominal aortic aneurysm will grow in the future in the MRA image. In this case, by using the first growth estimation model M1, it is possible to obtain an estimation result of whether or not abdominal aortic aneurysm is growing from an abdominal MRA image.

[0041] (Embodiment 2) In this embodiment, in addition to the configuration of Embodiment 1, an information processing device that estimates whether a cerebral aneurysm (unruptured cerebral aneurysm) occurring in a blood vessel in the brain of a subject will grow in the future based on the subject's subject information (patient information, clinical information). The information processing device 10 of this embodiment estimates whether a cerebral aneurysm will grow based on MRA images, as in Embodiment 1, and also estimates whether a cerebral aneurysm will grow based on the subject information, integrating the two estimation results to obtain a final estimation result (integrated growth information). The subject information includes, for example, attribute information such as the subject's (examinee's) age and gender, whether or not the subject has been diagnosed with hypertension (whether or not hypertension exists), whether or not the subject has been diagnosed with hyperlipidemia (whether or not hyperlipidemia exists), whether or not the subject has been diagnosed with diabetes (whether or not diabetes exists), smoking history, family history of subarachnoid hemorrhage (family (relatives') medical history of subarachnoid hemorrhage), size of the cerebral aneurysm, location of the cerebral aneurysm, presence or absence of a bleb (abnormal shape), UCAS-Japan score, and a number of other information. In addition to the above information, the subject information may also include vital data such as the subject's body temperature and blood pressure, various test results, medical history, medication history, etc. This information is stored for each patient (subject) in the electronic medical record data of the electronic medical record server, for example.

[0042] The information processing device of this embodiment has the same configuration as the information processing device 10 of embodiment 1, and therefore a description of the configuration will be omitted. Note that the information processing device 10 of this embodiment stores a second increase estimation model M2 in the storage unit 12 in addition to the configuration shown in Fig. 1. The second increase estimation model M2 is expected to be used as a program module constituting artificial intelligence software.

[0043] FIG. 6 is an explanatory diagram showing an example of the configuration of the second growth estimation model M2. The second growth estimation model M2 (third learning model) in FIG. 6 is trained to use subject information as input data, perform a calculation to determine (estimate) whether the subject's cerebral aneurysm will grow in the future based on the input subject information, and output the calculation result (second growth information regarding the presence or absence of cerebral aneurysm growth). The second growth estimation model M2 is configured using, for example, XGBoost, LightGBM, or CatBoost based on a Gradient Boosting Decision Tree (GBDT). Note that the second growth estimation model M2 is not limited to a decision tree, and may be configured using algorithms such as an SVM (Support Vector Machine), a Bayesian network, a regression tree, a CNN, or a Transformer, or may be configured using a combination of multiple algorithms.

[0044] The second growth estimation model M2 has an input layer, an intermediate layer, and an output layer. The input layer has multiple input nodes, each associated with information to be input. Each piece of subject information is input to the second growth estimation model M2 via its associated input node. The intermediate layer calculates output values ​​from each piece of information input via the input layer using various functions, thresholds, etc., and outputs the calculated output values ​​to the output layer. The output layer, like the first growth estimation model M1 of FIG. 2A , has two output nodes. One output node outputs the probability (certainty) of determining that the cerebral aneurysm of the subject in the input subject information will grow in the future, and the other output node outputs the probability (certainty) of determining that the cerebral aneurysm of the subject in the input subject information will not grow in the future. With the above-described configuration, when subject information is input, the second growth estimation model M2 outputs information (second growth information) indicating whether the cerebral aneurysm of the subject in the input subject information will grow in the future. In addition, the second augmentation estimation model M2 may be configured to weight each piece of input information according to its importance, for example, to weight heavily the presence or absence of blebs, the UCAS-Japan score, etc.

[0045] When using the second growth estimation model M2 of FIG. 6 , the information processing device 10 identifies the output node that outputs the largest output value (certainty factor) among the output values ​​from each output node, and identifies the discrimination result associated with the identified output node as the discrimination result to be estimated. The second growth estimation model M2 may also be configured to have a single output node that outputs a discrimination result with a high confidence factor (information indicating whether or not cerebral aneurysm growth is occurring) instead of two output nodes that output confidence factors for each discrimination result. The second growth estimation model M2 may also be configured to have a single output node that outputs the possibility of cerebral aneurysm growth in the subject of the input subject information, like the first growth estimation model M1a shown in FIG. 2B .

[0046] The second growth estimation model M2 is generated by machine learning using training data that associates training subject information (subject information on subjects with cerebral aneurysms) with correct labels indicating whether the subject's cerebral aneurysm will grow in the future. For example, the training data is generated by assigning a label (e.g., 1) indicating cerebral aneurysm growth to subject information on subjects who were diagnosed with a cerebral aneurysm and whose cerebral aneurysm grew during follow-up observation over several years to 10 years (a follow-up period), and by assigning a label (e.g., 0) indicating no cerebral aneurysm growth to subject information on subjects whose cerebral aneurysms did not grow during follow-up observation. The training data generated in this manner is stored, for example, in a training DB provided in the storage unit 12 and used during the learning process.

[0047] When subject information included in the training data is input, the second growth estimation model M2 learns so that the output value from the output node corresponding to the discrimination result indicated by the correct label approaches 1 and the output value from the other output node approaches 0. During the learning process, the second growth estimation model M2 performs calculations based on the input subject information to calculate output values ​​from each output node. The second growth estimation model M2 then compares the calculated output values ​​of each output node with values ​​corresponding to the correct label and optimizes parameters used in the calculation process so that the two values ​​approximate each other. The second growth estimation model M2 optimizes parameters such as weights (coupling coefficients) between nodes using backpropagation, steepest descent, or the like. This results in a second growth estimation model M2 that, when subject information is input, estimates whether the subject's cerebral aneurysm will grow in the future and outputs the estimation result.

[0048] The second growth estimation model M2 is not limited to the configuration shown in FIG. 6 . For example, the subject information input to the second growth estimation model M2 is not limited to the information shown in FIG. 6 , and at least two of the pieces of information shown in FIG. 6 may be input. Furthermore, in addition to or instead of the information shown in FIG. 6 , information or numerical values ​​of items that may affect the growth and rupture of cerebral aneurysms may be input. Furthermore, the information output from the second growth estimation model M2 may be the possibility (probability) of cerebral aneurysm growth, the risk level of cerebral aneurysm growth, etc., similar to the first growth estimation model M1. Learning of the second growth estimation model M2 may also be performed by the information processing device 10 or another learning device.

[0049] Next, a process for estimating whether a subject's cerebral aneurysm will grow in the future using the first growth estimation model M1 and the second growth estimation model M2 will be described. Figure 7 is a flowchart showing an example of the process procedure for estimating cerebral aneurysm growth in the second embodiment. The process shown in Figure 7 is the process shown in Figure 4 with steps S31 to S33 added between steps S23 and S24. Explanations of the same steps as in Figure 4 will be omitted.

[0050] After processing step S23, the control unit 11 of the information processing device 10 of this embodiment acquires subject information of a subject, such as a patient (S31). The control unit 11 acquires the subject information of the subject in the MRA image acquired in step S21, for example, from an electronic medical record server. Based on the acquired subject information, the control unit 11 determines whether the subject's cerebral aneurysm will grow in the future (S32). Specifically, the control unit 11 inputs the subject information into a second growth estimation model M2 and determines whether the cerebral aneurysm will grow based on the output value from the second growth estimation model M2. In addition to determining whether the cerebral aneurysm will grow using the second growth estimation model M2, the control unit 11 may also perform rule-based processing. For example, for each item included in the subject information, a score indicating the possibility that the content of each item (e.g., a numerical value) will affect the growth of the cerebral aneurysm may be assigned. The possibility of cerebral aneurysm growth may be calculated based on the score corresponding to the content of each item (e.g., a numerical value), and the presence or absence of cerebral aneurysm growth may be determined based on the calculated possibility.

[0051] The control unit 11 determines a final determination result (integrated growth information) based on the determination result determined from the MRA image of the cerebral aneurysm region in step S23 and the determination result determined from the subject information in step S32 (S33). For example, the control unit 11 calculates the average of the confidence levels for the determination result of cerebral aneurysm growth output from the first growth estimation model M1 and the confidence levels for the determination result of cerebral aneurysm growth output from the second growth estimation model M2, and determines this as the final determination result (the probability that the cerebral aneurysm should be determined to grow in the future). The control unit 11 may also weight the confidence levels output from each model M1 and M2 and calculate a weighted average to determine the final determination result.

[0052] The control unit 11 then performs steps S24 and S25. This allows the estimated result of whether or not a cerebral aneurysm is enlarged to be presented via a determination result screen such as that shown in FIG. 5 . Therefore, in this embodiment, when a cerebral aneurysm is discovered, it is possible to estimate whether or not the cerebral aneurysm will enlarge in the future and present this to a physician, achieving the same effect as in the first embodiment. In the process shown in FIG. 7 , steps S21 to S23 and steps S31 to S33 may be executed in reverse order or in parallel. This embodiment may also be configured to present an assessment result based on MRA images and an assessment result based on subject information in addition to the final assessment result (estimated result). Alternatively, only steps S31 to S33 of FIG. 7 may be executed, and the assessment result based on subject information may be presented. In a configuration in which only the assessment result based on subject information is presented, the presence or absence of cerebral aneurysm enlargement is estimated from information obtained by interviewing the subject and subject information such as previously diagnosed symptoms. Therefore, an estimated result of whether or not a cerebral aneurysm is enlarged can be obtained even without MRA images. In this embodiment as well, the modifications described in the first embodiment can be applied as appropriate.

[0053] (Embodiment 3) In the above-described embodiment 2, the first growth estimation model M1 was used to obtain an estimation result of the presence or absence of cerebral aneurysm growth from a brain MRA image (cerebral aneurysm region), and the second growth estimation model M2 was used to obtain an estimation result of the presence or absence of cerebral aneurysm growth from subject information, and a final estimation result was determined from the two estimation results. In this embodiment, an information processing device that uses one model to estimate the presence or absence of cerebral aneurysm growth from a brain MRA image (cerebral aneurysm region) and subject information will be described. The information processing device of this embodiment has the same configuration as the information processing device 10 of embodiment 1, so a description of the configuration will be omitted. In addition to the configuration shown in FIG. 1 , the information processing device 10 of this embodiment stores a third growth estimation model M3 in the memory unit 12. The third growth estimation model M3 is expected to be used as a program module constituting artificial intelligence software.

[0054] FIG. 8A is an explanatory diagram showing an example of the configuration of a third growth estimation model M3, and FIG. 8B is an explanatory diagram showing an example of the configuration of a fourth growth estimation model M4. The fourth growth estimation model M4 shown in FIG. 8B is a modified version of the third growth estimation model M3. The third growth estimation model M3 in FIG. 8A is trained to use an MRA image of a subject's brain (cerebral aneurysm region) and subject information as input data, perform a calculation to determine (estimate) whether the subject's cerebral aneurysm will grow in the future based on the input data, and output the calculation result. Note that the MRA image input to the third growth estimation model M3 is the same as the MRA image (cerebral aneurysm region) input to the first growth estimation model M1 shown in FIG. 2A, and the subject information input to the third growth estimation model M3 is the same as the subject information input to the second growth estimation model M2 shown in FIG. 6.

[0055] The third growth estimation model M3 has an input layer, an intermediate layer, and an output layer M3c. The input layer has multiple input nodes, through which the subject's cerebral aneurysm region (MRA image) and subject information are input. The intermediate layer includes an intermediate layer M3a of the first growth estimation model M1 and an intermediate layer M3b of the second growth estimation model M2. The MRA image of the cerebral aneurysm region is input to the intermediate layer M3a of the first growth estimation model M1, and the subject information is input to the intermediate layer M3b of the second growth estimation model M2. The intermediate layers M3a and M3b calculate output values ​​from the input data and output them to the output layer M3c. Like the first growth estimation model M1 and the second growth estimation model M2, the output layer M3c has two output nodes, each outputting a probability (certainty) of whether the cerebral aneurysm will grow in the future (whether or not there will be growth). With the above-described configuration, when an MRA image (cerebral aneurysm region) and subject information are input, the third growth estimation model M3 outputs information (growth information) indicating whether the cerebral aneurysm in the MRA image will grow in the future. The third growth estimation model M3 may also be configured to have a single output node that outputs a discrimination result with a high degree of certainty, or may be configured to have a single output node that outputs the possibility of cerebral aneurysm growth, like the first growth estimation model M1a shown in FIG. 2B.

[0056] The third growth estimation model M3 is generated by machine learning using training data that associates training MRA images (cerebral aneurysm regions) and subject information with correct labels indicating whether the cerebral aneurysms in the MRA images will grow in the future. The training MRA images, subject information, and correct labels can be the same as the training MRA images, subject information, and correct labels of models M1 and M2. When the MRA images (cerebral aneurysm regions) and subject information included in the training data are input, the third growth estimation model M3 learns so that the output value from the output node corresponding to the discrimination result indicated by the correct label approaches 1 and the output value from the other output node approaches 0. During the learning process, the third growth estimation model M3 performs calculations based on the input MRA images in the intermediate layer M3a and calculations based on the input subject information in the intermediate layer M3b, and outputs output values ​​based on the respective calculation results from each output node in the output layer M3c. The third growth estimation model M3 then compares the calculated output values ​​of each output node with the values ​​corresponding to the correct label and optimizes the parameters used in the calculation process so that the two values ​​approximate each other. Again, the third growth estimation model M3 optimizes parameters such as the weights (coupling coefficients) between nodes using backpropagation or the like. This results in a third growth estimation model M3 that, when an MRA image including a cerebral aneurysm region and subject information are input, outputs an estimation result indicating whether the subject's cerebral aneurysm will grow in the future. The learning of the third growth estimation model M3 may also be performed by the information processing device 10 or another learning device.

[0057] The third growth estimation model M3 is not limited to the configuration shown in FIG. 8A . For example, the information output from the third growth estimation model M3 may include the possibility (probability) of cerebral aneurysm growth, the risk level of cerebral aneurysm growth, etc. Furthermore, the third growth estimation model M3 is not limited to a configuration using intermediate layers M3a and M3b of the first growth estimation model M1 and the second growth estimation model M2, but may be configured as a single model as shown in FIG. 8B . The fourth growth estimation model M4 of FIG. 8B may be configured using algorithms such as CNN, decision tree, SVM, Bayesian network, regression tree, Transformer, etc., or may be configured by combining multiple algorithms. The fourth growth estimation model M4 can be generated by machine learning using the same training data as the training data used to train the third growth estimation model M3. During the training process, the fourth growth estimation model M4 receives MRA images (cerebral aneurysm region) and subject information included in the training data, and performs calculations based on the input MRA images and subject information to calculate output values ​​from each output node. The fourth growth estimation model M4 then optimizes parameters such as weights (coupling coefficients) between nodes using the backpropagation algorithm or the like so that the calculated output value of each output node approximates the value corresponding to the correct label. This results in a fourth growth estimation model M4 that outputs an estimation result of whether or not the cerebral aneurysm will grow in the future when an MRA image including the cerebral aneurysm region and subject information are input.

[0058] Next, a process for estimating whether a subject's cerebral aneurysm will grow in the future using the third growth estimation model M3 will be described. Figure 9 is a flowchart showing an example of the process procedure for estimating cerebral aneurysm growth in embodiment 3. The process shown in Figure 9 is the process shown in Figure 4 with step S41 added between steps S22 and S23. Explanation of the same steps as in Figure 4 will be omitted.

[0059] After processing step S22, the control unit 11 of the information processing device 10 of this embodiment acquires subject information of the subject (S41). Step S41 is the same process as step S31 in FIG. 7. The control unit 11 determines whether or not the cerebral aneurysm has grown based on the cerebral aneurysm region and subject information (S23). Here, the control unit 11 inputs the cerebral aneurysm region and subject information into a third growth estimation model M3 and determines whether or not the cerebral aneurysm has grown based on the output value from the third growth estimation model M3. The control unit 11 then performs steps S24 and S25. As a result, in this embodiment as well, the estimation result of whether or not the cerebral aneurysm has grown can be presented via a determination result screen such as that shown in FIG. 5.

[0060] In the process shown in Figure 9, the order of steps S21 to S22 and S41 may be reversed, and steps S21 to S22 and step S41 may be executed in parallel. Furthermore, in step S23 in Figure 9, the control unit 11 may use the fourth growth estimation model M4 instead of the third growth estimation model M3 to determine whether or not the cerebral aneurysm has grown. The configuration of this embodiment is applicable to the above-described first and second embodiments, and similar effects can be obtained even when applied to the first and second embodiments. The modified examples described in the above-described first and second embodiments can also be applied to this embodiment.

[0061] (Embodiment 4) In the above-described embodiments 1 to 3, the MRA image to be inspected for the presence or absence of cerebral aneurysm growth was an image in which a cerebral aneurysm was found. In this embodiment, an information processing device will be described that first detects a cerebral aneurysm from a brain MRA image and, if a cerebral aneurysm is detected, estimates the presence or absence of cerebral aneurysm growth from the cerebral aneurysm region (brain MRA image). The information processing device of this embodiment has the same configuration as the information processing device 10 of embodiment 1, and therefore a description of the configuration will be omitted. In addition to the configuration shown in FIG. 1, the information processing device 10 of this embodiment stores a cerebral aneurysm detection model M5 in the storage unit 12. The cerebral aneurysm detection model M5 is expected to be used as a program module constituting artificial intelligence software.

[0062] 10 is an explanatory diagram showing an example of the configuration of the cerebral aneurysm detection model M5. The cerebral aneurysm detection model M5 (second learning model) is a model that recognizes cerebral aneurysm regions contained in input MRA images of the brain, and can classify cerebral aneurysm regions from other regions in the MRA images on a pixel-by-pixel basis using, for example, semantic segmentation. The cerebral aneurysm detection model M5 is configured using models such as U-Net, nnU-Net, U-Net++, and Swin U-Net. The cerebral aneurysm detection model M5 is not limited to U-Net, and may be configured using an image segmentation algorithm such as SegNet, FCN (Fully Convolutional Network), or PSPNet (Pyramid Scene Parsing Network), or may be configured using an object detection algorithm such as CNN, R-CNN, Fast R-CNN, SSD (Single Shot Multibox Detector), or YOLO (You Only Look Once), or may be configured using a combination of multiple algorithms.

[0063] The cerebral aneurysm detection model M5 is trained to use an MRA image of the brain as input data, perform calculations to recognize cerebral aneurysm regions contained in the MRA image based on the input MRA image, and output information indicating the recognition results. Specifically, the cerebral aneurysm detection model M5 classifies each pixel of the input MRA image into a cerebral aneurysm region and other regions, and outputs a classified MRA image (hereinafter referred to as a labeled image) in which each pixel is associated with a label for each region. The labeled image output from the cerebral aneurysm detection model M5 can, for example, be an image in which pixels classified as a cerebral aneurysm region are represented by white pixels and pixels classified as other regions are represented by black pixels. In the example of Figure 10, the labeled image (white pixels classified as a cerebral aneurysm region) is superimposed on the MRA image of the input data to clearly show the cerebral aneurysm region indicated by the labeled image.

[0064] The cerebral aneurysm detection model M5 has an input layer, an intermediate layer, and an output layer. The input layer receives an MRA image of the brain to be processed. The intermediate layer includes a convolutional layer, a pooling layer, and a deconvolutional layer. The MRA image is input to the intermediate layer via the input layer, and the convolutional layer extracts image features by filtering or other processes to generate a feature map. The generated feature map is then compressed in the pooling layer. The deconvolutional layer enlarges (maps) the feature map generated by the convolutional layer and pooling layer to the original image size. The deconvolutional layer identifies the location of the cerebral aneurysm region in the image on a pixel-by-pixel basis based on the features extracted by the convolutional layer, and generates a labeled image by labeling each pixel as either a cerebral aneurysm region or another region.

[0065] The cerebral aneurysm detection model M5 is generated by machine learning using training data that associates training MRA images with correct labeled images in which each pixel in the MRA image is labeled (annotated) with data indicating the cerebral aneurysm region to be identified. The correct labeled images are generated from annotation images in which a cerebral aneurysm region or another region is assigned to each pixel in the MRA image by a medical expert. When an MRA image included in the training data is input, the cerebral aneurysm detection model M5 learns to output the correct labeled image included in the training data. Specifically, the cerebral aneurysm detection model M5 performs calculations in the intermediate layer based on the input MRA image and obtains a detection result for the cerebral aneurysm region detected in the MRA image. More specifically, the cerebral aneurysm detection model M5 acquires as output a labeled image in which each pixel in an MRA image is labeled with a value indicating the cerebral aneurysm region or another region. The cerebral aneurysm detection model M5 then compares the acquired detection result (labeled image) with the range of the cerebral aneurysm in the correct labeled image and optimizes parameters such as the weights (coupling coefficients) between nodes using an error backpropagation algorithm or the like so that the two are similar. This results in the cerebral aneurysm detection model M5 being able to output a labeled image indicating the cerebral aneurysm region in the input image when an MRA image of the brain is input. The training of the cerebral aneurysm detection model M5 may also be performed by the information processing device 10 or another learning device.

[0066] Next, a process for estimating whether a subject's cerebral aneurysm will grow in the future using the cerebral aneurysm detection model M5 and the first growth estimation model M1 will be described. Figure 11 is a flowchart showing an example of the process procedure for estimating cerebral aneurysm growth in embodiment 4, and Figure 12 is an explanatory diagram showing an example screen. The process shown in Figure 11 is the process shown in Figure 4 with steps S51 to S52 added between steps S21 and S22. Explanation of the same steps as in Figure 4 will be omitted.

[0067] After processing step S21, the control unit 11 of the information processing device 10 of this embodiment identifies (detects) a cerebral aneurysm region in the acquired MRA image of the brain (S51). Here, the control unit 11 inputs the MRA image of the brain into the cerebral aneurysm detection model M5 to perform segmentation, and identifies the cerebral aneurysm region in the MRA image based on the labeled image output from the cerebral aneurysm detection model M5. The control unit 11 calculates the maximum diameter of the cerebral aneurysm based on the cerebral aneurysm region identified in the MRA image (S52). The maximum diameter of the cerebral aneurysm can be calculated from the coordinate values ​​of each pixel in the cerebral aneurysm region in the MRA image.

[0068] Next, the control unit 11 extracts the cerebral aneurysm region from the MRA image based on the identified cerebral aneurysm region (S22). Here, the control unit 11 extracts a cerebral aneurysm region of a predetermined size (e.g., 64 pixels x 64 pixels x 32 pixels) using the center of gravity (center) of the identified cerebral aneurysm region in the MRA image. The control unit 11 then performs steps S23 to S25. In step S23, the control unit 11 inputs the extracted cerebral aneurysm region into the first growth estimation model M1 shown in FIG. 2A and determines whether or not the cerebral aneurysm has grown based on the output value from the first growth estimation model M1. In step S25, the control unit 11 displays a determination result screen, such as that shown in FIG. 12, on the display unit 15, thereby presenting the cerebral aneurysm region detected by segmentation in the MRA image, the location and maximum diameter of the cerebral aneurysm, as well as the estimated result of whether or not the cerebral aneurysm has grown. That is, the control unit 11 can present the location and maximum diameter of the cerebral aneurysm, and the possibility (probability) of cerebral aneurysm growth. In the above-described process, if a cerebral aneurysm region cannot be detected in step S51, the control unit 11 may skip steps S52 and S22 to S23 and store and display the determination result that no cerebral aneurysm is present.

[0069] The process of FIG. 11 is a process in which the configuration of this embodiment is applied to the information processing device 10 of embodiment 1, and cerebral aneurysm growth is estimated using the cerebral aneurysm detection model M5 and the first growth estimation model M1. The configuration of this embodiment can also be applied to the information processing device 10 of embodiments 2 and 3. Below, we will explain the process in which the configuration of this embodiment is applied to the information processing device 10 of embodiment 3, and cerebral aneurysm growth is estimated using the cerebral aneurysm detection model M5 and the third growth estimation model M3 or the fourth growth estimation model M4. FIG. 13 is a flowchart showing another example of the processing procedure for estimating cerebral aneurysm growth in embodiment 4. The process shown in FIG. 13 is the process shown in FIG. 9, with steps S61 to S62 added between steps S21 and S22. Steps S61 to S62 are the same as steps S51 to S52 in FIG. 11.

[0070] In the process of FIG. 13 , after step S21, the control unit 11 performs steps S61-S62, which are similar to steps S51-S52 of FIG. 11 , and then performs step S22. This allows the cerebral aneurysm region to be detected from the MRA image by segmentation, and based on the detected cerebral aneurysm region, the cerebral aneurysm region is extracted and input into the third growth estimation model M3 or the fourth growth estimation model M4. Next, the control unit 11 acquires the subject's subject information (S41), inputs the extracted cerebral aneurysm region and the acquired subject information into model M3 or M4, and determines whether or not the cerebral aneurysm has grown based on the output value from model M3 or M4 (S23). The control unit 11 then performs steps S24-S25. In step S25, the control unit 11 displays the determination result screen shown in FIG. 12 , thereby presenting the location and maximum diameter of the cerebral aneurysm in addition to the estimation result regarding whether or not the cerebral aneurysm has grown.

[0071] When the configuration of this embodiment is applied to the information processing device 10 of embodiment 2, the processing of steps S51 to S52 in FIG. 11 (steps S61 to S62 in FIG. 13) can be added between steps S21 and S22 in the processing shown in FIG. 7. The configuration of this embodiment can also achieve the same effect when applied to the above-mentioned embodiments 1 to 3. The modified examples described as appropriate in the above-mentioned embodiments 1 to 3 can also be applied to this embodiment.

[0072] (Embodiment 5) A modification of Embodiments 2 to 4 will be described. The second growth estimation model M2 of Embodiment 2 shown in FIG. 6 can be configured using explainable AI (XAI) and may be configured to input subject information of a subject and output a confidence level regarding the presence or absence of cerebral aneurysm growth and explanatory data indicating the basis for the estimation (prediction). This second growth estimation model M2 is configured to calculate, for each input data, a contribution (SHAP, Attention, etc.) that should serve as the basis for obtaining an estimation result (model output) using a SHAP or Attention mechanism, etc. The contribution is a value indicating the degree to which each input data of the second growth estimation model M2 contributed to the output of the estimation result. For example, SHAP is a value that quantitatively indicates the influence of each input data (input element) on the estimation result. Therefore, input data with a high contribution can be considered the basis for obtaining the estimation result. In addition to the configuration shown in FIG. 6, this second growth estimation model M2 has multiple output nodes (the number of input data) corresponding to each input data as output nodes for explanatory data. Therefore, the second augmented estimation model M2 can output the contribution (explanation data) to each input data from each output node for explanation data.

[0073] 14A and 14B are explanatory diagrams showing example screens. When using the second augmentation estimation model M2 configured as described above, the control unit 11 of the information processing device 10 can execute processing similar to that shown in FIG. 7. After processing step S32, the control unit 11 generates estimation basis information for notifying the input data that forms the basis of the estimation result identified in step S32. For example, the control unit 11 selects a predetermined number (e.g., three) of output values ​​(contribution degrees) from each output node for the explanatory data in descending order, and identifies the input data associated with the output node that output the selected output values ​​as the basis of the estimation. For example, if the UCAS-Japan score, the presence or absence of a bleb, and the size of a cerebral aneurysm are identified as input data that form the basis of the estimation result, the control unit 11 generates a message notifying the doctor of these basis information. The control unit 11 then displays the generated message on the judgment result screen as shown in FIG. 14A and presents it to a doctor or other professional. The screen of FIG. 14A displays the items (input data) that should be the basis for obtaining the estimation result (here, there is a high possibility of cerebral aneurysm enlargement) in addition to the configuration of the screen of FIG.

[0074] FIG. 14B is a modified example of the screen shown in FIG. 12. When the third growth estimation model M3 shown in FIG. 8A or the fourth growth estimation model M4 shown in FIG. 8B is configured using explainable AI, the models M3 and M4 input the subject's cerebral aneurysm region and subject information, and output a confidence level for the presence or absence of cerebral aneurysm growth and explanatory data indicating the basis for the estimation (the degree of contribution of the subject information to the input data). In this case, the control unit 11 can generate and present a determination result screen such as that shown in FIG. 14B. In addition to the configuration of the screen shown in FIG. 12, the screen shown in FIG. 14B displays items that should serve as the basis for the estimation result (input data of the subject information). With this configuration, in this embodiment, in addition to the estimation result of the presence or absence of cerebral aneurysm growth based on the MRA image and subject information, information on the basis for the estimation result can be presented. The basis for the estimation result presented in this manner can be used when formulating a treatment plan.

[0075] The effects of the above-described first to fourth embodiments are described below. FIGS. 15A to 15C are explanatory diagrams illustrating the effects of using the information processing device 10 of the first to fourth embodiments. The graph in FIG. 15A shows specificity on the horizontal axis and sensitivity on the vertical axis, and illustrates a receiver operating characteristic (ROC) curve based on the results of estimating the presence or absence of cerebral aneurysm growth from the existing UCAS-Japan score (ELAPSS score). The AUC (area under curve) of the ROC curve in FIG. 15A was calculated, resulting in an AUC of 0.667. The graph in FIG. 15B shows specificity on the horizontal axis and sensitivity on the vertical axis, and illustrates an ROC curve based on the results of estimating the presence or absence of cerebral aneurysm growth using the first growth estimation model M1. The first growth estimation model M1 in FIG. 15B was trained using MRA images from a total of 579 cases, including 125 cases in which cerebral aneurysms grew and 454 cases in which they did not. The ROC curve in Figure 15B yielded an AUC of 0.794. The graph in Figure 15C, with specificity on the horizontal axis and sensitivity on the vertical axis, shows an ROC curve based on the results of estimating the presence or absence of cerebral aneurysm growth using the second growth estimation model M2, and the ROC curve in Figure 15C yielded an AUC of 0.76. The results in Figures 15A to 15C show that the first growth estimation model M1 or the second growth estimation model M2 of the present disclosure can estimate cerebral aneurysm growth with higher accuracy than the conventionally used UCAS-Japan score.

[0076] Although the ROC curve is not shown, the ROC curve based on the results of estimating the presence or absence of cerebral aneurysm enlargement using the third enlargement estimation model M3 shown in Figure 8A yielded an AUC of 0.838, a sensitivity of 0.933, and a specificity of 0.684. In other words, it was found that the process of estimating the presence or absence of cerebral aneurysm enlargement based on MRA images and subject information can obtain more accurate estimation results than estimation processes based only on MRA images or only on subject information.

[0077] In the above-described first to fifth embodiments, the training data used to train models M1 to M4 were MRA images of individuals who were diagnosed with a cerebral aneurysm and whose cerebral aneurysms grew during a follow-up period of several years to approximately 10 years. Alternatively, training data may be generated based on MRA images of individuals whose cerebral aneurysms grew after a predetermined period, such as three or five years, since the discovery of the cerebral aneurysm. By learning using such training data, it is possible to generate a growth estimation model that can estimate whether a cerebral aneurysm will grow after a predetermined period (such as three or five years) based on the input MRA image (cerebral aneurysm region). Using such a model, when a cerebral aneurysm is discovered, it is possible to estimate and present not only whether a cerebral aneurysm will grow in the future (the likelihood of growth), but also whether a cerebral aneurysm will grow after a predetermined period, such as three or five years.

[0078] The matters described in each embodiment can be combined with each other. Furthermore, the independent claims and dependent claims described in the claims can be combined with each other in any and all combinations, regardless of the reference format. Furthermore, the claims use a format in which a claim references two or more other claims (multiple claim format), but this is not limited to this. A multiple claim (multi-multi claim) that references at least one other multiple claim may also be used.

[0079] The embodiments disclosed herein are to be considered in all respects as illustrative and not restrictive. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the meaning and scope of the claims.

[0080] REFERENCE SIGNS LIST 10 Information processing device 11 Control unit 12 Storage unit 13 Communication unit 14 Input unit 15 Display unit M1 First growth estimation model M2 Second growth estimation model M3 Third growth estimation model M4 Fourth growth estimation model M5 Cerebral aneurysm detection model

Claims

1. A program that causes a computer to execute a process of acquiring MRA (Magnetic Resonance Angiography) images of the brain, and acquiring enlargement information in the input MRA images by inputting the acquired MRA images of the brain into a learning model that outputs enlargement information regarding the presence or absence of enlargement of cerebral aneurysms in the input MRA images when the MRA images of the brain are input.

2. The program according to claim 1, which causes the computer to execute a process of inputting an acquired MRA image of the brain into a second learning model that recognizes cerebral aneurysms in the MRA image when the MRA image of the brain is input, thereby obtaining information indicating the recognized cerebral aneurysms.

3. The program according to claim 2, wherein the second learning model outputs an image obtained by segmenting the cerebral aneurysm region in the input MRA image.

4. The program according to claim 3, which causes the computer to execute a process of calculating the maximum diameter of the cerebral aneurysm based on an image obtained by segmenting the region of the cerebral aneurysm in the MRA image.

5. A program according to any one of claims 1 to 4, which causes the computer to execute the following process: extracting an area containing a cerebral aneurysm from the acquired MRA image; and inputting the extracted area containing a cerebral aneurysm into the learning model to obtain growth information for the cerebral aneurysm.

6. A program described in any one of claims 1 to 4 that causes the computer to execute the following process: acquire subject information about a subject; acquire second enlargement information by inputting the acquired subject information into a third learning model that outputs second enlargement information regarding whether or not cerebral aneurysms have enlarged when the subject information is input; and identify integrated enlargement information for the subject's cerebral aneurysm based on the acquired enlargement information and the second enlargement information.

7. The learning model is trained to output growth information regarding the presence or absence of growth of the subject's cerebral aneurysm when an MRA image of the subject's brain and subject information are input, and the program described in any one of claims 1 to 4 causes the computer to execute the following process: acquire subject information regarding the subject; and acquire growth information regarding the subject's cerebral aneurysm by inputting the acquired MRA image of the subject's brain and subject information into the learning model.

8. The program of claim 2, wherein the learning model is trained to output growth information regarding the presence or absence of growth of a cerebral aneurysm in a subject when an MRA image of the subject's brain and subject information are input, and the program causes the computer to execute the following processes: extracting an area of ​​a cerebral aneurysm in the MRA image based on information indicating a cerebral aneurysm recognized in the acquired MRA image; acquiring subject information regarding the subject; and acquiring growth information regarding the cerebral aneurysm in the subject by inputting the extracted area including the cerebral aneurysm and the acquired subject information into the learning model.

9. The program described in claim 6, wherein the subject information includes at least two of the subject's age, sex, presence or absence of hypertension, presence or absence of hyperlipidemia, presence or absence of diabetes, smoking history, family history of subarachnoid hemorrhage, size and location of cerebral aneurysm, presence or absence of bleb (irregular shape), and UCAS-Japan score.

10. A program according to any one of claims 1 to 4, wherein the growth information includes information regarding growth, which indicates that the size of the cerebral aneurysm will grow by more than a predetermined value after a predetermined period of time, and information regarding no growth, which indicates that the size will not grow by more than a predetermined value after a predetermined period of time.

11. A program according to any one of claims 1 to 4, which causes the computer to execute a process of displaying on a display unit the possibility of growth of the cerebral aneurysm in the MRA image based on the acquired growth information.

12. The program according to claim 3, which causes the computer to execute a process of displaying on a display unit the results of segmenting the area of ​​a cerebral aneurysm recognized in the acquired MRA image.

13. The program described in claim 6 causes the computer to execute a process in which the subject information includes multiple pieces of information, and information corresponding to the contribution of each piece of information included in the subject information input to the third learning model to the output of the second augmented information is obtained from the third learning model, and based on the obtained information, the information included in the subject information is displayed on a display unit in order according to the contribution to the output of the second augmented information.

14. An information processing method in which a computer executes a process of acquiring an MRA image of the brain, and acquiring information on growth in the input MRA image by inputting the acquired MRA image of the brain into a learning model that outputs growth information regarding the presence or absence of growth of cerebral aneurysms in the input MRA image when the MRA image of the brain is input.

15. An information processing device having a control unit, wherein the control unit acquires an MRA image of the brain, and acquires growth information in the input MRA image by inputting the acquired MRA image of the brain into a learning model that outputs growth information regarding whether or not cerebral aneurysms have grown in the input MRA image when the MRA image of the brain is input.

16. A model generation method in which a computer executes the process of acquiring training data including MRA images of the brain containing cerebral aneurysms and growth information regarding the presence or absence of growth of cerebral aneurysms in the MRA images, and using the acquired training data to generate a learning model that outputs growth information regarding the presence or absence of growth of cerebral aneurysms in the input MRA images when the brain MRA images are input.

17. A model generation method in which a computer executes the following process: acquiring training data including an MRA image of a subject's brain, subject information about the subject, and growth information regarding the presence or absence of growth of cerebral aneurysms in the MRA image; and using the acquired training data, generating a learning model that outputs growth information regarding the presence or absence of growth of cerebral aneurysms in the subject when the MRA image of the subject's brain and subject information are input.

Images