Brain image analysis device, brain image analysis system, and brain image analysis program

Abstract

Provided is a brain image analysis device that can provide information for assisting in the diagnosis of a plurality of neurodegenerative diseases. The brain image analysis device 1 receives an input of a brain tomographic image of an evaluation subject captured by a tomographic image scanner device, and three-dimensionally analyzes the degree of atrophy of a plurality of regions of interest in the brain. A segmentation unit 12 segments the whole brain tomographic image into structural units, and thus the structural unit related to the region of interest is identified among the segmented structural units. A brain structure volume signal intensity calculation unit 35 calculates the brain volume of each structural unit. A same age ratio calculation unit 16 integrates and analyzes the plurality of structural units for each age in the database, and analyzes the degree of atrophy of a prescribed region of interest in the brain by calculating an index value.

Description

Brain image analysis device, brain image analysis system, and brain image analysis program

[0001] The present invention relates to a brain image analysis device, a brain image analysis system, and a brain image analysis program for evaluating the degree of brain atrophy using brain image data measured by the brain image analysis device.

[0002] There are various types of neurodegenerative diseases, and doctors differentiate them by integrating information such as symptoms, various tests, and medical images.

[0003] Furthermore, since there are treatments available for some neurodegenerative diseases, such as those caused by hydrocephalus and metabolic diseases, correct differential diagnosis is important.

[0004] Dementia is one of these "neurodegenerative diseases," and it has become one of the most serious social issues in an aging society.

[0005] In this situation, for the diagnosis and differentiation of dementia, there have been reported techniques that attempt to assist diagnosis by using so-called biomarkers based on biochemical substances (see Patent Document 1) and evaluating motor function (see Patent Document 2). In the technique disclosed in Patent Document 2, a test for evaluating brain function is conducted by having the subject repeatedly point with a pointing unit (e.g., one's finger) at an index displayed on a display device a predetermined number of times while the subject's line of sight is not corrected (the line of sight is not shifted), and also repeatedly point with a pointing unit (e.g., one's finger) at an index displayed on a display device a predetermined number of times while the line of sight is corrected (the line of sight is shifted).

[0006] On the other hand, although there have been reports of the development of new therapeutic drugs (see Patent Document 3) and the exploration of new treatment methods such as ultrasound therapy (see Patent Document 4), it cannot be said that a fundamental treatment method for Alzheimer's disease has been established.

[0007] For this reason, current treatments are symptomatic treatments aimed at slowing the progression of symptoms. However, if the signs of dementia are noticed early and treatment is started at an early stage, it is possible to slow the progression while symptoms are still mild.

[0008] For example, according to the medical journal "Lancet" (see Non-Patent Document 1), it is known that as people age, the susceptibility to dementia changes due to the influence of past lifestyle habits (see Non-Patent Document 1). For example, education (1.6 times), high blood pressure (1.6 times), excessive drinking (1.2 times), obesity (1.6 times), smoking (1.6 times), depression (1.9 times), social isolation (1.6 times), lack of exercise (1.4 times), and diabetes (1.5 times).

[0009] Furthermore, according to the Japanese Society of Neurology's "2017 Dementia Disease Treatment Guidelines" (see Non-Patent Document 2), risk factors for lifestyle-related diseases such as hypertension, diabetes, and dyslipidemia can be risk factors that accelerate the progression of mild cognitive impairment to dementia, and therefore guidance on improving lifestyle habits and managing medication is necessary.

[0010] Therefore, at present, since dementia is thought to be closely related to lifestyle-related diseases, it is believed that detecting the signs as early as possible and reducing the risk of lifestyle-related diseases will indirectly lead to the prevention of Alzheimer's disease.

[0011] Specifically, it is believed that incorporating improvements to diet and regular exercise into daily life can help prevent Alzheimer's disease.

[0012] However, the most common early symptom of dementia is memory loss. However, as we all age, our memory deteriorates, so there are many cases where the person or their family does not notice the signs of dementia. And by the time they realize that "something is wrong," dementia has already progressed, and they may not be able to live without the support of those around them.

[0013] Medical imaging such as MRI (Magnetic Resonance Imaging) and CT (Computed Tomography) can non-invasively visualize brain structures, and these techniques can provide important information for the differentiation of some neurodegenerative diseases.

[0014] Therefore, Patent Document 5 discloses a system that indicates the risk of dementia by using the value of hippocampal volume (hippocampal volume) as brain state data of a subject imaged with an MRI (Magnetic Resonance Imaging) device.

[0015] Japanese Patent Publication No. 2020-144147 Japanese Patent No. 6218286 Japanese Patent No. 5033868 Japanese Patent Publication No. 2023-009246 Japanese Patent Publication No. 2021-097988

[0016] Deborah E Barnes, Kristine Yaffe, “The projected effect of risk factor reduction on Alzheimer's disease prevalence”, Lancet Neurol 2011; 10: 819-28, www.thelancet.com / neurology Vol 13 August 2014https: / / www.neurology-jp.org / guidelinem / nintisyo_2017.html

[0017] However, conventional dementia risk assessments have not necessarily been sufficient to objectively assess the risk of developing dementia from an early stage. Furthermore, conventional dementia risk assessments using MRI imaging have problems in that the volume of the hippocampus accounts for less than 1% of the total brain volume, and therefore not only is there a lack of accuracy in the assessment, but the amount of data accumulated is also insufficient due to the need for high-resolution imaging, which is limited by the availability of MRI devices capable of such imaging.

[0018] For example, Alzheimer's disease is considered to progress very slowly among dementia types, and it is said that in many cases, it takes about 20 years for the full-blown symptoms to appear after the onset of premonition symptoms. Therefore, there are currently no established methods for early detection.

[0019] The present invention has been made to solve such problems, and its purpose is to provide a brain image analysis device, a brain image analysis system, and a brain image analysis program that can provide information to assist in the diagnosis of multiple neurodegenerative diseases while utilizing brain tomographic image data accumulated as big data.

[0020] In order to achieve the above object, one aspect of the present invention is as follows. (Item 1)

[0021] According to one aspect of the present invention, a brain image analysis device receives a brain tomographic image of a subject captured by a tomographic scanner and three-dimensionally analyzes the degree of atrophy of multiple regions of interest in the brain, and includes a segmentation unit that performs segmentation by dividing the entire brain tomographic image into structural units, an identification unit that identifies structural units related to the region of interest from the divided structural units, a volume calculation unit that calculates the brain volume of each identified structural unit, a database containing age-related change data including brain volume data for each structural unit at each age under various imaging conditions for multiple subjects, and a comparison calculation unit that calculates an index value indicating a comparison of the volume of the structural unit of the subject to be evaluated with the value for the same age from the data for each structural unit at each age in the database, and the comparison calculation unit analyzes the degree of atrophy of a specified region of interest in the brain by integrating and analyzing the multiple structural units and calculating an index value. (Item 2)

[0022] Preferably, in the configuration of (Item 1), the index values ​​corresponding to atrophy of the plurality of regions of interest are linear combinations of values ​​obtained by standardizing the volumes of the corresponding structural units using the average volume and standard deviation of the same age group.

[0023] Preferably, in the configuration of (Item 1) or (Item 2), each combination of the degree of atrophy of a plurality of regions of interest is associated in advance with a plurality of disease types, and the contrast calculation unit analyzes the degree of atrophy of a predetermined region of interest in the brain in the integrated analysis to provide auxiliary information for diagnosis of a plurality of disease types.

[0024] Preferably, in any one of the configurations of (Item 1) to (Item 3), a display control means is further provided for displaying the degree of atrophy of the plurality of regions of interest as screen information, and the display control means generates information for displaying index values ​​corresponding to atrophy of the plurality of regions of interest as a spider chart for comparison with values ​​for people of the same age. (Item 5)

[0025] Preferably, in any one of (Item 1) to (Item 4), the plurality of regions of interest include the frontal lobe, the parietal lobe, the occipital lobe, the temporal lobe, and the brainstem.

[0026] Preferably, in the configuration of (Item 5), the comparison calculation unit calculates index values ​​using the anterior horn of the lateral ventricle as a substitute index for frontal lobe atrophy, the posterior horn of the lateral ventricle as a substitute index for parietal and occipital lobe atrophy, the inferior horn of the lateral ventricle as a substitute index for temporal lobe atrophy, and the third and fourth ventricles as substitute indexes for brainstem atrophy, and the display control means displays a spider chart showing each substitute index as a vertex in comparison with the corresponding value for the same age group. (Item 7)

[0027] Preferably, in any one of the configurations of (Item 1) to (Item 6), a display control means is further provided for displaying the degree of atrophy of the multiple regions of interest as screen information, and the display control means generates information to be displayed as a graph that expresses an index value showing a comparison with values ​​for people of the same age based on axes corresponding to the multiple regions of interest. (Item 8)

[0028] Preferably, in any one of the configurations of (Item 1) to (Item 7), the plurality of structural units include the hippocampus and the whole brain volume, and the contrast calculation unit i) generates data that displays the hippocampus volume on a first axis and the whole brain volume on a second axis that is perpendicular to the first axis, and ii) analyzes the degree of atrophy of a predetermined region of interest in the brain depending on which quadrant of the coordinates represented by the first axis and the second axis the calculated value of the brain volume of the subject belongs to. (Item 9)

[0029] Preferably, in any one of the configurations of (Item 1) to (Item 8), a structural unit selection unit is further provided that selects a structural unit that is related to atrophy of the region of interest and that changes less than other structural units due to the influence of imaging conditions based on brain volume data stored in the database, and the identification unit targets the structural unit selected by the structural unit selection unit for identification. (Item 10)

[0030] According to another aspect of the present invention, there is provided a brain image analysis system that receives input of brain tomographic images of a subject captured by a tomographic scanner and three-dimensionally analyzes the degree of atrophy of multiple regions of interest in the brain, the system comprising: a data server that stores a database of age-related change data, including brain volume data for each structural unit at each age under various imaging conditions, for multiple subjects; and a brain image analysis server that analyzes the degree of atrophy of the multiple regions of interest. The brain image analysis server includes a segmentation unit that performs segmentation by dividing the entire brain tomographic image into each structural unit, an identification unit that identifies a structural unit related to the region of interest from among the divided structural units, a volume calculation unit that calculates the brain volume of each identified structural unit, and a comparison calculation unit that calculates an index value indicating a comparison of the volume of the structural unit of the subject to be evaluated with that of the same age, based on the data for each structural unit at each age in the database. The comparison calculation unit integrates and analyzes the multiple structural units at each age in the database, and calculates an index value to analyze the degree of atrophy of the specified brain region of interest. (Item 11)

[0031] Preferably, in the configuration of (Item 10), the index values ​​corresponding to atrophy of the plurality of regions of interest are linear combinations of values ​​obtained by standardizing the volumes of the corresponding structural units using the average volume and standard deviation of the same age group.

[0032] Preferably, in the configuration of (Item 10) or (Item 11), each combination of the degree of atrophy of a plurality of regions of interest is associated in advance with a plurality of disease types, and the contrast calculation unit analyzes the degree of atrophy of a predetermined region of interest in the brain in the integrated analysis to provide auxiliary information for diagnosis of a plurality of disease types.

[0033] According to yet another aspect of the present invention, there is provided a brain image analysis program for causing a computer having a storage device and a computing device to execute processing as an analysis device that receives input of brain tomographic images of a subject captured by a tomographic image scanner and three-dimensionally analyzes the degree of atrophy of a predetermined region of interest in the brain, the storage device comprising a database containing age-related change data obtained by previously capturing images of a plurality of subjects and including brain volume data for each structural unit at each age under various imaging conditions, the computing device comprising the steps of: segmenting the entire brain tomographic image into each structural unit; identifying structural units associated with the region of interest from the segmented structural units; calculating the brain volume of each identified structural unit; and calculating an index value indicating a comparison of the volume of the structural unit of the subject to be evaluated with that of a person of the same age from the data for each structural unit at each age in the database, the step of calculating the index value including a step of integrating and analyzing the plurality of structural units at each age in the database to calculate an index value, thereby analyzing the degree of atrophy of the predetermined region of interest in the brain (Item 14).

[0034] Preferably, in the configuration of (Item 13), the index values ​​corresponding to atrophy of the plurality of regions of interest are linear combinations of values ​​obtained by standardizing the volumes of the corresponding structural units using the average volume and standard deviation of the same age group.

[0035] Preferably, in the configuration of (Item 13) or (Item 14), each combination of the degree of atrophy of a plurality of regions of interest is associated in advance with a plurality of disease types, and the step of calculating the index value includes a step of analyzing the degree of atrophy of a predetermined region of interest in the brain in the integrated analysis to provide auxiliary information for diagnosis of a plurality of disease types.

[0036] According to the present invention, it is possible to provide information that aids in the diagnosis of a number of neurodegenerative diseases while utilizing brain tomographic image data that has been accumulated as big data.

[0037] 1 is a functional block diagram of a brain image analysis device of this embodiment. FIG. 1 is a block diagram for explaining the hardware configuration of the brain image analysis device 1. FIG. 1 is a flowchart showing a pre-analysis of brain image data analysis processing. FIG. 2 is a flowchart showing analysis of a subject's brain image data analysis processing by the brain image analysis device. FIG. 2 is a diagram showing segmentation training data at various structural levels. FIG. 3 is a diagram showing age-related changes in the average value of ventricular volume and a predetermined variance value. FIG. 3 is a diagram showing the concept of a method for integrating structural units of the brain. FIG. 4 is a diagram for explaining the concept of segmenting the brain into structural units and combining them into larger structures. FIG. 4 is a diagram showing a flowchart of processing for evaluating atrophy of brain structures using composite indices. FIG. 5 is a diagram showing areas of the brain related to the frontal lobe. FIG. 6 is a diagram explaining an example of displaying morphological features of the brain. FIG. 7 is a diagram explaining an example of displaying morphological features of the brain. FIG. 8 is a diagram explaining another example of displaying morphological features of the brain.

[0038] Hereinafter, a system for utilizing brain image analysis data according to an embodiment of the present invention will be described.

[0039] In this embodiment, the brain image analysis device (or brain image analysis server) functions as a device that receives and analyzes brain image data. This brain image analysis device utilizes AI technology from Johns Hopkins University in the U.S. and uses software developed to analyze big data on 30,000 MRI brain images stored in Japan using AI to comprehensively evaluate cerebral atrophy and vascular changes, which are characteristics observed in patients with dementia and cerebral infarction.

[0040] Such brain image analysis technology developed by Johns Hopkins University in the United States is disclosed in the following documents, for example: Reference 1: U.S. Patent No. 10,074,173 Reference 2: U.S. Patent No. 10,535,133

[0041] In particular, Reference 1 discloses an automatic segmentation technique that can flexibly change the granularity level of the whole brain based on the hierarchical relationship of 254 structures defined in a brain atlas. Reference 2 discloses a technique that performs automatic segmentation of brain images and labels each structure using a generative model technique based on multiple atlases.

[0042] However, other segmentation methods and other labeling techniques may be used for segmenting brain images and labeling each structure.

[0043] In this specification, a "structural unit" of a brain image is described as referring to each of a plurality of structures (e.g., 505 structures) that have been automatically segmented based on one or more known brain atlases (brain maps), for example, using the method defined in Reference 1. However, as will be described later, the number of structures to be segmented is not limited to 505 structures, and for example, several segments may be integrated and considered as one "structural unit."

[0044] However, it is known that the structure of the brain is hierarchical and that its functions are localized, and the "structural units" obtained from brain images are not limited to those obtained using the method in Reference 1, but other divisions that can be distinguished from the information of the acquired images depending on the performance, type, imaging principle, imaging conditions, etc. of the scanner may also be used as "structural units."

[0045] As will be described later, a combination of multiple automatically segmented "structural units" may correspond to a single "brain structure." For example, the "brain structure" called the "ventricle" may consist of the lateral ventricle, the third ventricle, and the fourth ventricle, and the lateral ventricle may further have a hierarchical relationship consisting of an anterior portion, a posterior portion, and a lateral portion.

[0046] Hereinafter, "structural units" and "brain structures" formed by combinations of such structural units will be collectively referred to as "brain structures."

[0047] The brain image analysis device of this embodiment utilizes the above-described technology to quantify and evaluate ventricular expansion and the volume of the cerebral cortex in each part of the brain, and outputs reports such as the following: 1) "Comprehensive brain atrophy assessment" that evaluates the aging progression of the degree of brain atrophy by comparing with subjects of the same age.

[0048] 2) "Vascular lesion assessment" involves quantifying the volume of white matter lesion areas and comparing them with patients of the same age to assess the progression of vascular lesions in the brain over time.

[0049] The program that runs on the brain image analysis device is a program for brain checkups and health centers that uses AI analysis of lesions or deviations from the average based on these evaluations to measure analytical values ​​and obtain analytical results that serve as auxiliary or reference information for diagnosing brain health risks and diseases. <Principle of the analysis device>

[0050] First, the technical background and principles of the brain image analysis device will be explained.

[0051] Brain image analysis devices enable the extraction of information related to dementia by automatically quantifying the degree of brain atrophy and vascular lesions from MRI images. Conventional brain checkups have used brain MRI to detect and diagnose aneurysms, brain tumors, cerebral infarction, and intracerebral inflammation. However, they have not yet been able to predict the risk of developing dementia as a lifestyle-related disease and help reduce or prevent the onset of dementia. While existing technology includes programs for measuring the volume of the hippocampus, which plays a major role in cognitive function, the hippocampus accounts for only 0.3% of the human brain's volume. Volumetric measurements of the brain's parenchymal structure are prone to variability depending on the type of MRI equipment, resolution, and imaging method, making it difficult to accumulate useful information as stable tracking data. Furthermore, as discussed below, much of the remaining 99.7% of brain structure is important for estimating dementia risk, limiting the value of measuring the hippocampus alone.

[0052] Furthermore, measuring hippocampal atrophy requires a high-resolution brain MRI, but taking MRI images takes time, so high-resolution images are often not taken during brain checkups. Brain checkups in Japan have a long history of over 20 years, with over 1 million pieces of data already in existence, and some locations even having time-series data going back more than 15 years. However, for the reasons mentioned above, they are not necessarily able to reduce the risk of developing dementia or address risk management for dementia, which will become increasingly important in the future.

[0053] Analysis using the brain image analysis device of this embodiment makes it possible to accumulate large amounts of big data and perform automatic AI analysis, which will greatly contribute to future dementia risk management. Research using this brain image analysis device is already underway using existing data from brain checkup centers and research institutions. This analysis is a technology that is expected to lead to world-class big data analysis in the future, with existing data alone exceeding one million items.

[0054] By using this brain image data analysis, it is possible to improve the stability of measurements by measuring structures that are less affected by MRI image acquisition conditions, such as ventricular volume, and that sensitively reflect brain atrophy, and to realize a brain image analysis data utilization system that can more effectively utilize the results of analyzing brain atrophy data from brain checkup data. <Configuration>

[0055] 1 is a functional block diagram of a brain image analysis device according to this embodiment. As will be described later, the brain image analysis device 1 also includes a configuration for setting parameters, conditions, and the like for analysis in advance based on existing big data in order to analyze the subject currently being evaluated.

[0056] As shown in FIG. 1 , the brain image analysis device 1 of embodiment 1 includes an image input unit 11, a segmentation unit 12, a brain structure volume / brain signal intensity calculation unit 31, an imaging condition calibration unit 15, a peer ratio calculation unit 16, a cerebrovascular disorder degree calculation unit 17, an analysis report creation unit 18, and an analysis report output unit 19.

[0057] The brain structure volume / brain signal intensity calculation unit 31 includes a brain structure selection unit 32 , an imaging condition influence evaluation unit 33 , a brain structure identification unit 34 , and a brain structure volume / signal intensity calculation unit 35 .

[0058] The "image input unit 11" accepts input of brain images (brain MRI images) acquired by an MRI scanner or brain images and imaging condition data transmitted from a PACS (Picture Archiving and Communication System), as well as attribute data of the imaged subject. The image input unit 11 corresponds to a communication interface that allows the brain image analysis device 1, which is realized as a computer, to exchange data with the outside. The PACS may be a server on the cloud, and the brain image analysis device itself may also be a server on the cloud.

[0059] The "segmentation unit 12" performs segmentation to divide the entire brain MRI image into structural units. Although not particularly limited, this segmentation can utilize the techniques described in the above-mentioned references 1 and 2, for example.

[0060] The 'imaging condition calibration unit 15' performs calibration in accordance with the imaging conditions when an image is captured by the MRI scanner, in order to minimize the influence of differences in the imaging conditions.

[0061] The 'peer age ratio calculation unit 16' calculates the peer age ratio of each structural unit of each subject and an index value (described later) from the structural unit data for each age in the database 2.

[0062] The 'cerebrovascular disorder degree calculation unit 17' calculates the degree of cerebrovascular disorder.

[0063] The analysis report creation unit 18 creates an analysis report based on the analysis results obtained.

[0064] The database 2a is a big data database and includes brain volume data for each structural unit at each age under various imaging conditions (including resolution, slice thickness, slice angle, slice direction, and sequence parameters (echo time, repetition time, flip angle, etc.)), age-structural unit volume data including brain signal intensity data, and age-structural unit signal intensity data. This database may be provided within an analysis center, or may be installed on a cloud server. Note that the data referred to as "big data" here may include not only existing data captured and integrated at multiple imaging sites, but also data that is newly acquired at multiple imaging sites, including non-existent ones, and that is accumulated sequentially.

[0065] As described above, the brain image analysis device 1 is realized by executing arithmetic processing on a computer. Therefore, the segmentation unit 12, the brain structure volume / brain signal intensity calculation unit 31, the imaging condition calibration unit 15, the peer ratio calculation unit 16, the cerebrovascular disorder degree calculation unit 17, and the analysis report creation unit 18 correspond to functions executed by a CPU (Central Processing Unit) based on a program stored in memory, and this program includes modules that execute each function. In addition, the database 2a can also store information to be included in a report corresponding to the analysis results.

[0066] The imaging condition calibration unit 15 performs calibration using the age-structural unit volume data and age-structural unit signal intensity data in the database 2a. After calibration, the same-age ratio calculation unit 16 calculates the same-age ratio and index value of each structural unit of each subject from the structural unit data for each age in the database 2, thereby analyzing the risk of developing dementia and the possibility of disease, and obtaining auxiliary information or reference information.

[0067] The brain structure volume / signal intensity calculation unit 31 will be further described below.

[0068] As described above, the brain structure volume / signal intensity calculation unit 31 corresponds to a function in which the CPU executes a corresponding module in a program stored in memory when the computer operates as the brain image analysis device 1. The brain structure selection unit 32, the imaging condition influence evaluation unit 33, the brain structure identification unit 34, and the brain structure volume / signal intensity calculation unit 35 are implemented as submodules in the module of the brain structure volume / signal intensity calculation unit 31, each of whose functions is executed by the CPU.

[0069] The "brain structure volume / signal intensity calculation unit 31" first includes an imaging condition influence evaluation unit 33 that evaluates the influence of the imaging conditions of brain MRI images on brain structures based on big data before analyzing the current subject, and a brain structure selection unit 32 that, based on the evaluation results, selects brain structures that are highly effective in terms of being able to accurately calculate the volume of brain structures, etc. from an engineering perspective even if there are variations in imaging conditions and subject attributes (age, gender, etc.), and that exhibit large changes due to aging and disease. The "brain structure volume / signal intensity calculation unit 31" further includes a brain structure identification unit 34 that identifies brain structures that are less affected by the imaging conditions from the imaging data from the subject currently being evaluated, and a brain structure volume / signal intensity calculation unit 35 that calculates the volume and signal intensity of large structures, such as ventricles and lobar structures, from among the identified brain structures.

[0070] FIG. 2 is a block diagram for explaining the hardware configuration of the brain image analysis device 1 shown in FIG.

[0071] In the following description, the brain image analysis device 1 is assumed to operate on a physical server 3000 that operates on the cloud. However, the brain image analysis device 1 may operate on an on-premise server, or may be installed at the location of the device that captures brain images, such as a workstation.

[0072] As described above, the server 3000 may be configured such that a computing device (CPU: Central Processing Unit) within its own housing executes the arithmetic processing, or such that part of the program processing is executed on another server. In the following description, it is assumed that the arithmetic device within its own housing executes the arithmetic processing.

[0073] Referring to FIG. 2, the server 3000 comprises a computer device 3010, a network communication unit 3300 for communicating with a network, and a recording medium (e.g., a memory card) 3210 for recording data from the outside and providing the data to the computer device 3010.

[0074] For example, a USB memory, a memory card, or an external storage device can be used as the recording medium 3210. Furthermore, for example, a wired LAN or wireless LAN communication function can be used as the network communication unit 3100.

[0075] As shown in FIG. 2 , the computer main body constituting this computer device 3010 includes, in addition to a disk drive 3030 and a memory drive 3020, a CPU (Central Processing Unit) 3040, each connected to a bus 3050, memory including a ROM (Read Only Memory) 2060 and a RAM (Random Access Memory) 3070, a nonvolatile rewritable storage device 3080, and an input / output interface 3090 for communication via a network and for sending and receiving data with external devices. The nonvolatile storage device 3080 may be, for example, a hard disk drive (HDD) or a solid state drive (SSD). The following description will be given assuming that the drive is an SSD. An optical disk can be inserted into the disk 3030. A memory card 2210 can be inserted into the memory drive 3020 .

[0076] When the programs of the computer device 3010 operate, the data and programs that store the information that is the basis for the operation of the computer will be described as being stored in the SSD 3080.

[0077] 2, the medium capable of recording information such as a program to be installed in the computer main body may be, for example, a DVD-ROM (Digital Versatile Disc), a memory card, a USB memory, etc. In such cases, the computer main body is provided with a drive device (memory drive 3020, disk drive 3030) capable of reading these media.

[0078] The main components of the computer device 3010 are computer hardware and software executed by the CPU 3040. Generally, such software is stored in a storage medium and distributed or distributed via a network, and is obtained via the disk drive 3030 or the network communication unit 3300 and temporarily stored in the SSD 3080. The software is then read from the SSD 3080 into the RAM 3070 in the memory and executed by the CPU 3040. Note that when connected to a network, the software may be directly loaded into the RAM and executed without being stored in the SSD 3080.

[0079] When distributed, the program for functioning as computer device 3010 does not necessarily have to include an operating system (OS) that causes computer main body 3010 to execute functions of an information processing device, etc. The program only needs to include an instruction portion that calls appropriate functions (modules) in a controlled manner and achieves the desired results. How computer system 3010 operates is well known, and a detailed description will be omitted.

[0080] Furthermore, the CPU 3040 may be a single-core processor or a multi-core processor. That is, it may be a single-core processor or a multi-core processor. The server 3000 may also be configured with multiple servers to execute distributed processing.

[0081] 3 is a flowchart showing a pre-analysis of the brain image data analysis process, and FIG. 4 is a flowchart showing the analysis of the brain image data analysis process for a subject by the brain image analysis device. (Pre-analysis using big data)

[0082] 3, before analysis of the current subject, the image input unit 11 executes an image input step of reading image data from the database 2a based on big data (step S501). The segmentation unit 12 executes a segmentation step for the image input in the image input step S501 (step S502). The segmentation in the segmentation step S502 divides the brain image into multiple structural units. Segmentation can also be performed by preparing, for example, training data that has already been segmented in the database 2a (which stores imaging condition data, age-structural unit volume data, and age-structural unit signal intensity data).

[0083] For example, by using training data such as that shown in US Pat. No. 10,074,173 in Reference 1, the structure of the brain can be defined at multiple structural levels.

[0084] Figure 5 shows segmentation training data at various structural levels. In the example of Figure 5, the coarseness (granularity) of the brain structure definition is defined in five levels: 287 structures, 137 structures, 54 structures, 19 structures, and 8 structures. Finer structure definitions increase the amount of information about brain regions, but reduce data reproducibility and increase dependency on imaging conditions. On the other hand, coarser structure definitions decrease the amount of information about brain regions, but increase data reproducibility and decrease dependency on imaging conditions. As a result of segmentation, all of the defined structures can serve as indicators (volume values, signal intensity values) for measuring the brain's health status. The brain image analysis device 1 can select structures that are more suitable as health indicators from these many structures, and by performing preprocessing, it is possible to obtain a final indicator with high suitability.

[0085] Here, the correspondence between the "structural units" and the "brain structures" used in the analysis is as follows:

[0086] That is, for example, segmentation can be performed in the smallest unit, and a larger structure can be defined by combining multiple units.

[0087] 3, next, the brain structure volume / signal intensity calculation unit 31 executes an imaging condition influence evaluation step S503 and a brain structure selection step S504. Each processing step executed by the brain structure volume / signal intensity calculation unit 31 will be described in detail below.

[0088] First, as a module of the imaging condition influence evaluation program, the imaging condition influence evaluation unit 33 executes an imaging condition influence evaluation step (step S503). The imaging condition influence evaluation step S503 is a step for evaluating the influence of various imaging conditions from an engineering perspective. The imaging condition influence evaluation unit 33 selects structures that are highly robust from an engineering perspective and evaluates them based on the data in the database 2a. In this step S503, the influence of brain MRI images taken under various conditions on the definition of the structure (e.g., volume) is evaluated. This is because, for example, when MRI images are taken at different resolutions, the volume of a specific structure may be measured as being larger.

[0089] For example, there are nearly 100 parameters in the MRI imaging conditions, and it cannot be assumed that the imaging conditions are the same between imaging sites (e.g., hospitals). All of these parameters have the potential to affect the volume.

[0090] As a result, strictly speaking, it is impossible to compare data between facilities as it is, and each hospital must collect hundreds or thousands of data points before it is possible to make an assessment such as, "The average hippocampal volume of a 50-year-old is XX, and this person's hippocampus is smaller in comparison." This process must be repeated even if one imaging condition is changed.

[0091] Therefore, when, for example, 505 brain structures are divided by segmentation, the imaging condition influence evaluation unit 33 executes a process of evaluating and separating those that are sensitive to differences in these parameters, with their volume values ​​changing when the parameters are changed, from those that are insensitive, with their volume values ​​remaining almost unchanged regardless of the parameters. The latter is an excellent index from an engineering perspective. Therefore, the imaging condition influence evaluation unit 33 evaluates "how much the volume values ​​change depending on the imaging parameters" based on, for example, changes in the CV (Coefficient of Variation) value or the volume-age correlation curve due to the imaging conditions.

[0092] Next, as a module of the brain structure selection program, the brain structure selection unit 32 executes a brain structure selection step (step S504). In the brain structure selection step S504, brain structures with high suitability for brain structure volume and signal calculation processing are selected. The brain structures selected here include structures with known medical and biological functions, or structures known to be medically important for the disease risk being evaluated. For example, the hippocampus, which is known to be important for memory. Furthermore, from an engineering perspective, structures suitable as indicators require different suitabilities, and in the brain structure selection step 504, brain structures with a large effect on variability from an engineering perspective are selected.

[0093] Below, we will explain the significance of "brain structures with a large effect despite variability." From an engineering perspective, the three most important suitability factors are good reproducibility (small variability when multiple measurements are taken under the same conditions), small influence from imaging conditions (robustness; similar values ​​are obtained even when measurements are taken under different conditions), and high sensitivity to the biological changes we want to detect. In this context, "sensitivity" refers to the ratio of the biological change (effect), its magnitude, and reproducibility; even if the effect is high, if the variability is large, it cannot be considered an excellent indicator from an engineering perspective.

[0094] Therefore, if the risk of dementia is to be evaluated, for example, the brain structure selection unit 32 will determine, from the standpoints of sensitivity and reproducibility, as will be described later, to use only the volume of the ventricles as an index from the 505 structures obtained by segmentation from the imaging data of the subject, or to use an index that is a combination of the ventricles and other brain structures. The brain structure selection unit 32 stores the brain structures selected in this way, for example, in database 2a, in association with the disease to be evaluated, as selected brain structure data.

[0095] The method of defining the brain structure at multiple levels using the training data described above, as shown in Figure 5, is extremely effective in finding robust structures.

[0096] Thereafter, the brain image analysis apparatus 1 of the first embodiment executes processing by the brain structure volume / signal intensity calculation unit 35 (S505) and processing by the imaging condition calibration unit 15 (step S506). <Processing by the imaging condition calibration unit 15>

[0097] The imaging condition calibration unit 15 first performs a "brain structure volume-age correlation creation process" in which, prior to analysis of the current subject, it creates a volume-age correlation curve for the measurement results of brain structures of multiple people of different ages (each subject) based on existing big data.

[0098] Next, the imaging condition calibration unit 15 performs a "brain structure volume-age correlation parameter extraction process" to extract, for each imaging condition, volume-age correlation parameters that define the correlation curve created by the brain structure volume-age correlation creation process, as well as statistical parameters (such as average values).

[0099] Next, the imaging condition calibration unit 15 performs an "imaging condition parameter matching process" to determine calibration coefficients to match or minimize the volume and signal intensity of each brain structure so that the differences in the parameters of the volume-age correlation curves and statistical parameters extracted in the "brain structure volume-age correlation parameter extraction process" are minimized even when the imaging conditions are different. Here, the "calibration coefficient" is not particularly limited, but refers to, for example, a coefficient used to calibrate the volume of a specific brain structure so that it matches between imaging conditions A and B. Similarly, the "calibration coefficient" refers to a coefficient used to calibrate the signal intensity so that it matches between imaging conditions A and B. Note that, for example, the "brain structure volume-age correlation parameter" can be the average value and standard deviation of the structure at each age. Therefore, calibration is a correction between group A and group B (for example, two groups imaged using MRI devices with different resolutions). If we assume that the average for Group A is 10% larger than the average for Group B at age 30, and Group A is considered the standard, then by reducing the volume of the 30-year-old patients in Group B by 10%, the data from Group B can be compared with the data from Group A.

[0100] To calibrate the difference in volume values ​​due to different shooting conditions, the shooting condition calibration unit 15 can also take images of the same person under different shooting conditions and measure the influence of each shooting condition (for example, resolution) to perform calibration. Steps S701 to S703 show one method of calibrating the difference in volume values ​​due to different shooting conditions (when not using the same person).

[0101] For example, there are two methods for determining the calibration coefficients:

[0102] In the first method, for example, if the average value and standard deviation of a 30-year-old person under condition A are average AA and deviation SA, and under condition B they are average AB and deviation SB, then by setting volume data B of the person photographed under condition B to B+(average AA-average AB), the average value of data B photographed under condition B can be compared with condition A. In other words, (average AA-average AB) becomes the calibration coefficient for B→A conversion. In this case, it is assumed that there is a certain amount of measurement data under conditions A and B, and that the averages of each can be calculated.

[0103] The statistical parameters of both conditions may be matched based on either condition A or condition B. For example, if the condition is the strength of the measurement magnetic field, the statistical parameters may be matched to the measurement data under the condition of the highest magnetic field strength, which is considered to provide the highest resolution. Alternatively, if the slice thicknesses under conditions A and B are different, the statistical parameters may be matched to the measurement data under the condition of the thinner slice thickness, which allows for the observation of a finer structure.

[0104] The second method is to use the statistical value of A as prior information when there is a large amount of data for condition A and a small amount of data for condition B.

[0105] A more rigorous implementation of this concept can be achieved using so-called Bayesian estimation.

[0106] In the above explanation, calibration is performed by comparing the average value and standard deviation for each age group and achieving a match. However, for example, when the average value is in the 20s to 80s range, the average values ​​for ages 20 to 80 under conditions A and B may be applied to a polynomial, and calibration coefficients may be calculated so that the polynomials match, even if the coefficients of the polynomials are used. In this case, differences in imaging conditions may also be added as covariates.

[0107] The above-mentioned Bayesian estimation will be explained in more detail below.

[0108] That is, for example, let us say that the average of structure X at age 30, found from 30,000 cases under imaging condition A, is θ, and the standard deviation is γ. Let us say that under imaging condition B, the results of volume measurement of structure X for n people are average μ and standard deviation σ. A Bayesian model can be used to estimate the posterior distribution. f(θ|data) ∝ f(θ)f(d|θ) For example, if both the prior distribution f(θ) and likelihood f(d|θ) follow normal distributions, the posterior distribution (estimation of the average under imaging condition B) is: Mean: θγ 2 / (nγ 2 +σ 2 ) + μnγ / (nγ 2 +σ 2 ) Variance: γ 2 σ 2 / (nγ 2 +σ 2 ) is a normal distribution. (Analysis for current subjects)

[0109] Next, with reference to FIG. 4, the process of analysis performed by the brain structure volume / signal intensity calculation unit 31 based on the brain image data and imaging condition data of the subject currently being evaluated will be described.

[0110] After the image input step (step S510) and the segmentation step (step S512), the brain structure volume / signal intensity calculation unit 31 executes a brain structure identification step S514, a brain structure volume / signal intensity calculation step S516, processing by the imaging condition calibration unit 15 (step S518), and processing by the same age ratio calculation unit 16 (step S520).

[0111] First, for analysis of the current subject, the image input unit 11 executes an image input step in which brain image data and imaging condition data of the subject to be evaluated are transmitted from the imaging site (the facility where the imaging device is installed and imaging is performed) (step S510). The segmentation unit 12 executes a segmentation step for the image input in the image input step S510 (step S512). The segmentation in the segmentation step S512 divides the brain image into multiple structural units (e.g., 505 structural units). Again, the segmentation can be performed with reference to the segmentation training data in the database 2a described above.

[0112] Next, as a module of the brain structure identification program, the brain structure identification unit 34 performs labeling of brain structures for the structural units divided by segmentation, and executes a brain structure identification step of identifying structures that are less affected by the imaging conditions according to the prior evaluation by the imaging condition influence evaluation unit 33 (step S514). In the brain structure identification step S514, structures are identified that are robust and less affected by the imaging conditions, and that can be minimized by calibration, and that are selected taking into account their sensitivity to the disease risk of the evaluation target. As described above, examples of robust structures include the ventricles and large lobe structures (e.g., the frontal lobe) when dementia risk is the target.

[0113] Next, as a module of the brain structure volume / signal intensity calculation program, the brain structure volume / signal intensity calculation unit 35 calculates the volume and signal intensity for each brain structure (step S516). That is, the volume and signal intensity of the brain structure are calculated from the structures with large volumes, such as the ventricles and lobe structures, identified in step S514.

[0114] Here, for example, the volume value of the identified brain structure is calculated by multiplying the number of voxels contained in the structure by the volume of one voxel. Furthermore, the signal intensity calculation is equivalent to the average value of the signal intensities of all voxels contained in the structure. In this case, the signal intensity calculation is not limited to the average value, and may be a median, moment, or histogram.

[0115] Here too, the method of defining the brain structure at multiple levels using the training data described above, as shown in FIG. 5, is extremely effective in finding robust structures.

[0116] Thereafter, the brain image analysis device 1 of the first embodiment executes the processing of the imaging condition calibration unit 15 (step S518) and the processing of the same age ratio calculation unit 16 (step S520). <Processing of the same age ratio calculation unit 16>

[0117] The peer ratio calculation unit 16 executes an age-brain volume distribution creation process, an age-brain volume distribution parameterization process, an age percentile calculation process, and an index value calculation process as described below.

[0118] Here, too, the peer age ratio calculation unit 16 corresponds to a function in which the CPU executes the corresponding module in the program stored in the memory when the computer operates as the brain image analysis device 1. The age-brain volume distribution creation process, age-brain volume distribution parameterization process, age percentile calculation process, and index value calculation process are implemented as sub-modules in the module of the peer age ratio calculation unit 16, each of which is executed by the CPU. <Configuration Description>

[0119] The age-based brain volume distribution generation process generates a population distribution of the volumes of selected brain structures at each age, where "population distribution" refers to the distribution of the mean and standard deviation of volumes obtained from three or more subjects.

[0120] The age-brain volume distribution parameterization process calculates parameters of the volume distribution of brain structures.

[0121] The age percentile calculation process uses the calculated parameters to calculate an age percentile for each subject based on their age.

[0122] The age-brain volume distribution generation process generates a population distribution of the volume of the selected structures at each age.

[0123] Next, the age-brain volume distribution parameterization process parameterizes the population distribution. Experimental results showed that using the logarithm of ventricular volume on the horizontal axis approaches a normal distribution. Therefore, for example, if a normal distribution model is used, the distribution can be parameterized using two values: the mean and standard deviation.

[0124] FIG. 6 is a diagram showing age-dependent changes in the mean value of ventricular volume and a predetermined variance.

[0125] As an alternative to the parameterization described above, where the horizontal axis represents the logarithm of the ventricular volume and the normal distribution is used, a method (e.g., Bayesian statistics) can be used to directly calculate percentiles from raw data, as shown by the dashed line in Figure 6.

[0126] Next, the age percentile calculation process can use the determined parameters to determine percentiles based on the age of each individual.

[0127] The cerebrovascular disorder degree calculation unit 17 calculates the degree of cerebrovascular disorder.

[0128] First, to calculate the volume of brain structures containing abnormal white matter signal regions, we incorporate the abnormal white matter signal regions into the segmentation training data.

[0129] The same age ratio is calculated from the age correlation of the brain white matter signal abnormality area in the database 2a.

[0130] Next, as a simultaneous calculation process for cerebral atrophy and cerebral white matter signal abnormality regions, the degree of cerebral atrophy and cerebrovascular disease are calculated simultaneously from the same MRI image.

[0131] As described above, the brain shape and degree of atrophy can be determined using the volume of each structure. Meanwhile, the degree of vascular damage in the brain can also be measured using the structural definitions set forth in the training data. For example, by defining abnormal signal intensity regions in the white matter of a specific brain structure using the training data, the volume of the abnormal signal regions can be automatically calculated for each individual in the volume calculation process for the abnormal brain white matter signal regions as part of the volume calculation process for the abnormal brain white matter signal regions. [Integrated analysis of structural units of the brain based on brain images]

[0132] As described above, cross-sectional images of the brain, such as MRI images and CT images, provide important information for the differentiation of some neurodegenerative diseases.

[0133] For example, hydrocephalus often causes significant dilation of the ventricles while narrowing of the sulci in the upper part of the brain.Also, typical Alzheimer's disease often shows significant atrophy of the medial temporal lobe, including the hippocampus.

[0134] For example, in frontotemporal dementia, significant atrophy is often seen in the frontal and temporal lobes.

[0135] Alternatively, for example, brain atrophy may not be significant in the early stages of Lewy body dementia or early-onset Alzheimer's disease.

[0136] Furthermore, for example, significant white matter lesions are often seen in vascular dementia.

[0137] Anatomical features obtained from these images provide important clues for differentiating various types of neurodegenerative diseases. However, except for typical and prominent cases, these features are difficult to distinguish in many patients, being "weak features" or "complex features." Their detection requires the physician's experience, and even with a great deal of experience, they are unclear. One of the reasons for this difficulty is that physician interpretation relies on subjective classification.

[0138] Therefore, tools that can objectively and quantitatively describe these characteristics are useful. However, it is difficult to characterize various types of neurological disorders with a single numerical value (e.g., the volume of a single structure). Therefore, a composite index is needed.

[0139] In the brain image analysis device of this embodiment, several types of neurodegenerative diseases are characterized by at least two indices, the volume of multiple brain structures and the image signal intensity (hereinafter referred to as "integrated analysis"), and the "likelihood" of each type (the likelihood of belonging to each type) is quantitatively evaluated. Note that such an index that represents the morphological characteristics of the brain by combining at least two indices will be called a "composite index." (Concept of the integrated analysis method of brain structural units)

[0140] FIG. 7 is a diagram showing the concept of the integrated analysis method for structural units of the brain.

[0141] For example, MRI images can be T1-weighted images, T2-weighted images, diffusion-weighted images, or derivatives of T2-weighted images such as FLAIR (Fluid Attenuated Inversion Recovery) images, which utilize the characteristics of T1 relaxation time and T2 relaxation time to visualize brain structure.

[0142] For example, T2-weighted images are indispensable for detecting lesions because they extract water as a high signal (appears white), and many lesions (inflammation / infarction, etc.) appear as high signals. On the other hand, FLAIR images are images in which normal water (cerebrospinal fluid in the skull) appears as a low signal (appears black) on T2-weighted images, and are considered useful for identifying periventricular lesions and lesions accompanied by necrosis / dropout (replacement by cerebrospinal fluid).

[0143] In contrast, in CT images, differences in X-ray transparency are displayed as shades of black and white, so bones and hard objects appear white, water, muscle, fat, etc. appear gray, and air appears black.

[0144] Therefore, in this embodiment, after taking an image of the brain using an MRI device or a CT scanner, an analysis is performed to characterize the shape of the brain.

[0145] Here, "characterizing the shape of the brain" means quantitatively determining whether a certain part is larger or smaller than the average brain shape for each age, and this can be illustrated using the method described below.

[0146] For example, the two figures shown in FIG. 7(a) and FIG. 7(b) belong to the higher concept of "circle" and have the same area.

[0147] If these are expressed as a single number (for example, area), the two circles (a perfect circle and an ellipse) can be characterized as having the same shape.

[0148] On the other hand, if it is possible to segment these circles into five regions as shown in FIG. 7, then the perfect circle and ellipse can each be represented by five areas.

[0149] In this case, if it were a perfect circle, the areas of regions 1, 2, 3, and 4 would all be the same, whereas with an ellipse, the following relationship holds: (area of ​​region 2) = (area of ​​region 3) < (area of ​​region 1) = (area of ​​region 4).

[0150] In other words, by using the areas of these five regions, the difference between a perfect circle and an ellipse can be expressed numerically.

[0151] Therefore, if we use this way of thinking, when we have MRI images of two brains (brain A and brain B), we can calculate the volume of each part of the brain and compare them. By using the volumes of each part, we can express how the shapes of brain A and brain B differ using the numbers representing the volume of each part.

[0152] In the following, the above-mentioned "structural unit" is used as such a portion. (Structural unit used for characterization)

[0153] However, assuming that the shape of the brain is characterized using the volume of each structural unit as described above, the next question to consider is, "What regions (structural units) should be defined and their volumes measured to most appropriately clarify the differences between Brain A and Brain B types?"

[0154] However, the "most appropriate region definition" depends on the application (in this case, medical purpose), so there is no single answer.

[0155] However, one of the medical purposes of the brain image analysis device of this embodiment is to differentiate between various subtypes of dementia that cause dementia (a condition in which cognitive function declines to the point where it interferes with daily life).

[0156] Of course, a final diagnosis of dementia subtype cannot be made solely from brain morphology using MRI images. While brain morphology is important and necessary information for differentiation, it does not satisfy sufficient conditions. However, classifying brain morphology into clusters can provide supplementary information for diagnosing dementia subtypes.

[0157] On the other hand, among each subtype, there are some that have "typical morphological characteristics," and whether or not a patient has these typical characteristics is important clinical information. The brain image analysis device of this embodiment quantitatively evaluates these "typical morphological characteristics" as described below.

[0158] The technology that enables such quantitative evaluation can be found in the following US patent technology, US Patent No. 10,074,173 "A Method to Analyze Human Anatomy from Multiple Granularity Levels," cited in the above-mentioned Reference 1.

[0159] The contents of this patent are also disclosed in the following paper: Reference 3: Aigerim Djamanakova, et al. "Tools for multiple granularity analysis of brain MRI data for individualized image analysis", NeuroImage 101 (2014) pp.168-176

[0160] These documents disclose a method for "feature extraction," which is the most important problem in analyzing an image and grasping its characteristics. In other words, they disclose a method for expressing important features in a small number of numerical values ​​(for example, 10 numerical values) from a brain image consisting of features of 1 million pixels (approximately 10 MB).

[0161] Here, "image analysis" in this embodiment refers to the process of deriving clinically useful features and interpretations from images containing a very large amount of information, which is nothing other than information compression. For example, if a diagnosis can be made from an image (or auxiliary information for a diagnosis can be obtained), this corresponds to converting the image into features containing less than 20 characters but containing a large amount of clinical information, such as "This patient has XX disease" ("XX in this patient indicates the possibility of XX disease").

[0162] There are two main methods for extracting features from such images or compressing information.

[0163] Method 1: If the captured image is a 3D image with a resolution of 1 mm, there will be 1.2 million pixels in a 1.2 liter brain. This is an overwhelming amount of information, considering that meaningful (required) compression in normal medical practice ultimately amounts to a few sentences or 10 numbers. One way to improve this is to increase the size of each pixel. For example, increasing the pixel size from 1 mm to 8 mm will result in 8 x 8 x 8 = 256 times the amount of information compression.

[0164] Method 2: The brain is divided (segmented) into several structural units, and larger structures are defined by gradually combining the defined structural units, ultimately redefining the brain into approximately 10 large structures, for example. Defining the brain into approximately 10 structures results in significant compression and increases clinical value.

[0165] FIG. 8 is a diagram illustrating the concept of segmenting the brain into structural units and combining them into larger structures.

[0166] In this embodiment, in accordance with the concept explained in FIG. 7, specifically, "Method 2" is adopted, which is a combination of segmentation and structural units as shown in FIG.

[0167] However, each defined structure becomes larger, reducing the ability to characterize the brain's shape in detail. In the example of the circle and ellipse in Figure 7, using a single number (total area) was not enough to describe the difference between the circle and ellipse. The degree of compression and the ability to characterize shape are generally inversely proportional.

[0168] Therefore, the above-mentioned "approximately 10 structural units" is merely an example, and this number will be changed as appropriate when clustering brain shapes into groups with desired characteristics.

[0169] In the present embodiment, the following description will discuss image feature extraction (information compression) using a method such as that shown in FIG. 8, with the aim of providing auxiliary diagnostic information for the differential diagnosis of dementia.

[0170] We will also explain how to display the compressed numerical value, which is usually about 10.

[0171] Hereinafter, the method of calculating diagnostic auxiliary information for dementia by such a "combination of multiple structural units" will be referred to as "evaluation of atrophy of brain structures using composite indices."

[0172] FIG. 9 is a flowchart showing a process for evaluating atrophy of brain structures using a composite index.

[0173] Referring to FIG. 9, the process of analysis performed by the brain structure volume / signal intensity calculation unit 31 based on the brain image data and imaging condition data of the subject currently being evaluated will be described.

[0174] After the image input step (step S810) and the segmentation step (step S812), the brain structure volume / signal intensity calculation unit 31 executes a brain structure identification step S814, a brain structure volume / signal intensity calculation step S816, and processing by the imaging condition calibration unit 15 (step S818).

[0175] Here, too, first, for the analysis of the current subject, the image input unit 11 executes an image input step in which it reads brain image data and imaging condition data of the subject to be evaluated, transmitted from the imaging site (step S810). The segmentation unit 12 executes a segmentation step for the image input in the image input step S810 (step S812). The segmentation in the segmentation step S812 divides the brain image into multiple structural units (e.g., 505 structural units). Again, the segmentation can be performed with reference to the segmentation training data in the database 2a described above.

[0176] Next, as a module of the brain structure identification program, the brain structure identification unit 34 performs brain structure labeling on the structural units divided by segmentation, and executes a brain structure identification step of identifying structures previously selected to be used in calculating a composite index according to the prior evaluation by the imaging condition influence evaluation unit 33 (step S814). In the brain structure identification step S814, as will be described later, at least two brain structures (structural units) are identified. Note that, although not particularly limited, at least one of the brain structures should be a robust structure that is less affected by imaging conditions and can be minimized by calibration, and should preferably be a structure such as a ventricle or a nearby ventricle selected in consideration of sensitivity to the disease risk of the evaluation target.

[0177] Next, as a module of the brain structure volume / signal intensity calculation program, the brain structure volume / signal intensity calculation unit 35 executes calculation of the volume and signal intensity for each identified brain structure (step S816).

[0178] Then, in the brain image analysis device 1 of embodiment 2, after the imaging condition calibration unit 15 performs processing (step S510), the same-age ratio calculation unit 16 calculates a composite index using multiple brain structures (S820), and further performs an integrated analysis, as described below, on the volumes of at least two brain structures used in calculating the composite index, generates display data for displaying a graph showing the results (S822), and outputs it as report information.

[0179] As explained above, when the brain is divided into multiple regions and each region is evaluated, it is found that each region exhibits characteristic age-related changes. For example, there is almost no age-related change in hippocampal volume until the age of 60, but the frontal lobe begins to atrophy in the 30s. By understanding normal age-related changes in each region of the brain, it is possible to detect pathological atrophy localized to a particular region. Therefore, by knowing the atrophy of multiple regions in combination, it is more likely that we can identify the signs of a specific disease (for example, in addition to the above, frontotemporal dementia causes atrophy in a combination of the frontal lobe and temporal lobe).

[0180] Therefore, by combining information on multiple structures in addition to the total brain volume and ventricular volume, it is possible to more accurately depict the state of brain atrophy. This increases the accuracy of determining the brain condition to be evaluated (risk of dementia, dementia class, causative diseases of dementia, etc.) (Method of identifying structures and compressing information)

[0181] The example shown in FIG. 8 illustrates a method for segmenting the brain in four stages, with 254, 60, 21, and 11 structures defined in each stage.

[0182] The brain image analysis device of this embodiment performs segmentation having at least one stage.

[0183] For example, 254 structures are defined at the leftmost side of Figure 8. If all four stages are used, 254 + 60 + 21 + 11 = 346 structures are defined.

[0184] From these, segments with known characteristics of typical cases of each type of dementia will be identified and selected.

[0185] For example, in neurodegenerative diseases of the brain, the following characteristics are known for each structural unit: i) Hydrocephalus: 1) Enlargement of the lateral ventricles 2) Enlargement of the Sylvian sulcus 3) Shrinkage of the CSF area between the upper brain and the skull 4) Shrinkage of the sulci in the frontal and parietal lobes 5) Atrophy of the entire brain tissue

[0186] Here, the "CSF region" refers to the area where cerebrospinal fluid (CSF) is present. CSF is a colorless, transparent liquid that exists between the brain, spinal cord, and the membrane (dura mater) that surrounds them. It is produced in the ventricles (cavities within the brain), circulates, is absorbed on the surface of the brain, and returns to the veins. Its role is not completely clear, but it is thought to mainly regulate the water content of the brain and maintain its shape.

[0187] Therefore, by comparing the structures (structural units) themselves, or other structural units that correlate with the expansion or contraction of these structural units, with the average for the same age, it is possible to calculate whether the brain morphology of a typical hydrocephalic patient is maintained.

[0188] As an example, for volume comparison of structural units, a "Z score" is used to standardize the volume values, as defined below: Z score: (subject volume - age-matched mean volume) / age-matched volume standard deviation

[0189] As the index value, instead of a Z-score, the ratio of the subject's volume to the average volume of structural units of the same age (same-age ratio), or a normalized value (normalized value) obtained by dividing the deviation from the minimum value of the data for structural units of the same age (= the value when the minimum value is set to 0) by the data range (= maximum value - minimum value) may be used. In this specification, the process in which the same-age ratio calculation unit 16 calculates the above-mentioned Z-score, same-age ratio, or normalized value using the measured values ​​for subjects of the same age is referred to as a "comparison calculation for calculating an index." Note that the calculation is not limited to these, as long as it calculates an index for comparing the volume of the structural units of the subject with data of the same age.

[0190] Using such Z scores, for example, the following hydrocephalus type index can be created:

[0191] That is, for the five structures mentioned above, the sum of the volumes of the left and right structures is used for simplicity as a measure of the deformation (expansion / contraction) of the brain shape associated with hydrocephalus. If each structure (structural unit) is numbered 1-5, the index is expressed as the signed average of the Z scores for each structural unit, as follows:

[0192] Here, the signs of the Zscores of structures 1 and 2 and structures 3, 4, and 5 are reversed because structures 1 and 2 expand and structures 3, 4, and 5 contract.

[0193] That is, the stronger the tendency toward hydrocephalus, the larger the index is assigned.

[0194] Furthermore, if the importance of each structure differs, weighting can be performed as follows.

[0195] Dementia caused by neurodegenerative diseases is said to be of three types: Alzheimer's, Lewy body, and frontotemporal. In particular, Alzheimer's and frontotemporal have the following characteristics: ii) Dementia: Alzheimer's 1) Limbic system structures in the medial temporal lobe (hippocampus, amygdala) shrink 2) The inferior horn of the lateral ventricle enlarges 3) The entire brain shrinks iii) Dementia: Frontotemporal 1) The frontal lobe shrinks 2) The temporal lobe shrinks 3) The parietal and occipital lobes are relatively preserved

[0196] Furthermore, spinocerebellar degeneration, an incurable neurological disease, is a condition in which nerve cells from the cerebellum and brainstem to the spinal cord gradually decrease, causing ataxia. As the symptoms progress, the patient becomes unable to move their body as they wish, with symptoms including unsteadiness when walking, trembling hands, and slurred speech. This disease is said to have the following characteristics: iv) Spinocerebellar degeneration 1) Atrophy of the spinal cord and cerebellum 2) Enlargement of the fourth ventricle 3) Relative preservation of other areas

[0197] Vascular dementia is a type of dementia that occurs when nerve cells in the brain are destroyed by cerebrovascular disorders (strokes), such as cerebral infarction and cerebral hemorrhage. It is considered to be the second most common type of dementia overall. Since cerebrovascular disorders are a disease that develops suddenly, cognitive symptoms also appear suddenly. This is a major difference from neurodegenerative diseases. The following are the characteristics that appear on brain images: v) Vascular dementia 1) The volume of white matter lesions increases

[0198] By selecting brain structures (including structural units) that reflect these characteristics from the above segments, it is possible to numerically or graphically represent whether or not a particular case is typical of each type of dementia.

[0199] Also, the lack of atrophy of these structures is important information in that it reduces the possibility of the above type.

[0200] The above describes a configuration in which structures that may be important features from a medical point of view are selected in advance using existing medical knowledge.

[0201] However, when selecting such a structure, it is also necessary to choose the structure from an engineering perspective.

[0202] That is, for clinical doctors, since there are a wide variety of types of scanner devices and imaging conditions for capturing brain images, it is desirable to select a structure that is less susceptible to imaging conditions.

[0203] For example, the ventricles are known to be more robust in terms of measurement accuracy than the brain parenchyma. Also, the larger the structure, the higher the measurement reliability. For these reasons, it is desirable to select ventricles or large structures whenever possible.

[0204] For example, when defining a brain structure, consider the case of defining the frontal lobe.

[0205] FIG. 10 is a diagram showing the areas of the brain related to the frontal lobe.

[0206] There are multiple ways to define the frontal lobe.

[0207] The first method is to define the cortical area of ​​the frontal lobe, as shown by the white line in Figure 10(a). This is a common method, but the cortex is a structure that is only about 2-3 mm in size, and a deviation of a single pixel boundary can potentially change the size by 30-50%. Therefore, it is easily affected by imaging conditions and image resolution.

[0208] Therefore, a second method is envisioned in which the frontal lobe is defined as a larger unit by segmenting the cortex and white matter together, as shown by the white line in Figure 10(b).

[0209] Hereafter, this method will be referred to as the "simultaneous cortex-white matter segmentation method." If the "frontal lobe" is selected as the brain structure to be clinically evaluated in order to assess the degree of brain expansion or contraction (atrophy), and taking into account the engineering viewpoint mentioned above, the structure identified by the "simultaneous cortex-white matter segmentation method" would be preferable.

[0210] Furthermore, since the anterior horn of the lateral ventricle often enlarges when the frontal lobe atrophies, the influence of imaging conditions can be further reduced by using enlargement of the anterior horn of the lateral ventricle as a surrogate indicator of frontal lobe atrophy.

[0211] Furthermore, because hippocampal atrophy correlates with ventricular enlargement (the more the hippocampus atrophies, the more the ventricles enlarge), if the hippocampus is selected clinically as the brain structure to be evaluated, and taking into account the above-mentioned engineering viewpoint, it is also possible to use the volume of the ventricles as an alternative indicator by inverting its sign (diagram of evaluation results).

[0212] There are several ways to visualize the evaluation indicators obtained in this way, but here we will introduce two. <1: Spider Chart>

[0213] 11 and 12 are diagrams illustrating examples of displaying morphological features of the brain.

[0214] That is, Figures 11 and 12 extract and display morphological features of different types of dementia.

[0215] Here, seven brain structures were selected from the above-mentioned clinical and engineering viewpoints. The seven brain structures are as follows: 1) Anterior horn of the left lateral ventricle 2) Anterior horn of the right lateral ventricle 3) Posterior horn of the left lateral ventricle 4) Posterior horn of the right lateral ventricle 5) Inferior horn of the left lateral ventricle 6) Inferior horn of the right lateral ventricle 7) Third and fourth ventricles

[0216] Here, the index values ​​are calculated using the anterior horn of the lateral ventricle as a substitute indicator for frontal lobe atrophy, the posterior horn of the lateral ventricle as a substitute indicator for parietal and occipital lobe atrophy, the inferior horn of the lateral ventricle as a substitute indicator for temporal lobe atrophy, and the third and fourth ventricles as substitute indicators for brainstem atrophy.

[0217] Figure 11(a) is an example of a spider chart for the frontotemporal type. In the figure, the regular heptagon represents the average value of the subject's age (generation), and is shown in comparison with the subject's value.

[0218] As mentioned above, the frontotemporal type is characterized by "atrophy of the frontal lobe," "atrophy of the temporal lobe," and "relative preservation of the parietal lobe and occipital lobe." Therefore, as shown in Figure 11(a), the frontotemporal type is characterized by atrophy of the anterior horn of the lateral ventricle and the inferior horn of the lateral ventricle as an alternative indicator.

[0219] Figure 11(b) is an example of a spider chart for Alzheimer's disease. Here too, the regular heptagon has the same significance.

[0220] The Alzheimer's disease is characterized by "atrophy of limbic system structures in the medial temporal lobe (hippocampus, amygdala)," "enlargement of the inferior horn of the lateral ventricle," and "atrophy of the entire brain." Therefore, as shown in Figure 11(b), the Alzheimer's disease is characterized by atrophy of the inferior horn of the lateral ventricle as an alternative indicator.

[0221] Figure 12(c) is an example of a spider chart for spinocerebellar degeneration, where the regular heptagon has the same significance.

[0222] As mentioned above, spinocerebellar degeneration is characterized by "atrophy of the spinal cord and cerebellum," "enlargement of the fourth ventricle," and "relative preservation of other parts of the brain." Therefore, as shown in Figure 12(c), spinocerebellar degeneration is characterized by enlargement of the third and fourth ventricles.

[0223] By using a spider graph like the one shown above, it is possible to show whether a subject has the characteristics of each dementia type.

[0224] As mentioned above, by adjusting the average age of the subjects (80 years old in this case) to form a regular heptagon, the characteristics of each type can be made to stand out.

[0225] In this example, we created a spider graph by selecting seven axes based on the volume of the ventricles, but by adding other structures, we can see the characteristics of other types. For example, in the case of hydrocephalus, by adding the Sylvian sulcus (enlarged), the frontal lobe and parietal sulci (shrinking), and the parietal CSF area (shrinking), it is possible to illustrate whether the patient has typical morphological characteristics of hydrocephalus. <2: 2D and 3D diagrams>

[0226] FIG. 13 is a diagram illustrating another example of displaying morphological features of the brain.

[0227] FIG. 13 shows an example in which two indices are combined and displayed two-dimensionally.

[0228] However, for example, three indices may be combined and displayed three-dimensionally.

[0229] In the example of FIG. 13, two structures that exhibit volume changes characteristic of a particular disease are selected from the segmented structures.

[0230] For example, the volume of the whole brain and the volume of the hippocampus are shown in FIG. 13, which shows the ratio of each to the same age group.

[0231] If the two selected structures always undergo similar changes in, say, a certain type of dementia (for example, if the whole brain shrinks by 10%, the hippocampus also shrinks by 10%), the subject's evaluation index will be distributed as an open circle. On the other hand, if the two structures change independently, the distribution will be as black circles. An open circle indicates a correlation between whole brain and hippocampal volume, while a black circle indicates no correlation.

[0232] In the latter case, this two-dimensional diagram can be used to classify brains into four quadrant types based on their morphology: whole brain, non-hippocampal atrophy, whole brain atrophy, specific hippocampal atrophy, and hippocampal sparing. Depending on which quadrant a patient falls into, their brain morphology can be objectively characterized. For example, it is known that there are two types of Alzheimer's dementia: one with specific hippocampal atrophy and one with preserved hippocampus.

[0233] When selecting these two axes, it is effective to first select two structures that are not correlated, such as black circles. It is also possible to select them based on past clinical knowledge.

[0234] For example, hydrocephalus often causes significant dilation of the ventricles while narrowing of the sulci in the upper part of the brain. In this case, the ventricles and the sulci in the upper part of the brain can be selected as indicators. If these two are highly correlated, the brain parenchyma volume and the ventricles can be selected as two axes.

[0235] For example, typical Alzheimer's disease often shows significant atrophy of the medial temporal lobe, including the hippocampus.

[0236] In addition, frontotemporal dementia often shows significant atrophy of the frontal and temporal lobes.

[0237] Alternatively, brain atrophy may not be evident in the early stages of Lewy body dementia or early-onset Alzheimer's disease.

[0238] Furthermore, for example, significant white matter lesions are often seen in vascular dementia.

[0239] Based on this existing clinical knowledge, a combination of two axes can be envisioned.

[0240] The above explanation is about a method for selecting two axes based on knowledge, but it is also possible to select two or three axes based on data such as principal component analysis.

[0241] For example, the axis may be selected using a linear combination of the volumes of the structures using principal component analysis, as follows:

[0242] As described above, the brain image analysis device of this embodiment makes it possible to provide information that assists in diagnosis by utilizing brain tomographic image data accumulated as big data and performing analysis of multiple neurodegenerative diseases characterized by at least two or more of the volume of multiple brain structures and the image signal intensity.

[0243] The embodiments disclosed herein are merely examples of configurations for specifically implementing the present invention, and do not limit the technical scope of the present invention. The technical scope of the present invention is defined by the claims, not by the description of the embodiments, and is intended to include modifications within the literal scope of the claims and within the scope of equivalent meanings.

[0244] 1 Brain image analysis device, 2a Database, 11 Image input unit, 12 Segmentation unit, 31 Brain structure volume / brain signal intensity calculation unit, 15 Shooting condition calibration unit, 16 Peer age ratio calculation unit, 17 Cerebrovascular disorder degree calculation unit, 18 Analysis report creation unit 19 Analysis report output unit, 1000 Brain image analysis server.

Claims

1. A brain image analysis device that receives input of a brain tomographic image of a subject taken by a tomographic image scanner device and analyzes the degree of atrophy of multiple regions of interest in the brain in three dimensions, comprising: a segmentation unit that performs segmentation to divide the entire brain tomographic image into structural units; an identification unit that identifies structural units related to the region of interest from the divided structural units; a volume calculation unit that calculates the brain volume of each of the identified structural units; a database containing age-related change data including brain volume data of each of the structural units at each age under various shooting conditions, obtained by previously imaged images of multiple subjects; and a contrast calculation unit that calculates an index value indicating a comparison of the volume of the structural unit of the subject to be evaluated with the value of the same age, from the data of each of the structural units at each age in the database, wherein the contrast calculation unit analyzes the degree of atrophy of the specified region of interest in the brain by integrating and analyzing a plurality of the structural units and calculating the index value.

2. A brain image analysis device as described in claim 1, wherein the index values ​​corresponding to atrophy of the multiple regions of interest are each a linear combination of values ​​obtained by standardizing the volume of the corresponding structural units using the average volume and standard deviation for the same age group.

3. A brain image analysis device as described in claim 1 or 2, wherein each combination of the degree of atrophy of the multiple regions of interest is associated in advance with a multiple type of disease, and the contrast calculation unit analyzes the degree of atrophy of the specified region of interest in the brain in the integrated analysis to provide auxiliary information for diagnosis of the multiple types of disease.

4. A brain image analysis device as described in claim 3, further comprising a display control means for displaying the degree of atrophy of the multiple regions of interest as screen information, said display control means generating information for displaying index values ​​corresponding to atrophy of the multiple regions of interest as a spider chart for comparison with values ​​for the same age group.

5. A brain image analysis device as described in claim 4, wherein the multiple regions of interest include the frontal lobe, the parietal lobe, the occipital lobe, the temporal lobe, and the brain stem.

6. The brain image analysis device of claim 5, wherein the comparison calculation unit calculates index values ​​using the anterior horn of the lateral ventricle as a substitute indicator of frontal lobe atrophy, the posterior horn of the lateral ventricle as a substitute indicator of parietal and occipital lobe atrophy, the inferior horn of the lateral ventricle as a substitute indicator of temporal lobe atrophy, and the third and fourth ventricles as substitute indicators of brainstem atrophy, and the display control means displays a spider chart showing each of the substitute indicators as a vertex in comparison with the corresponding value for the same age group.

7. A brain image analysis device as described in claim 1, further comprising a display control means for displaying the degree of atrophy of the multiple regions of interest as screen information, wherein the display control means generates information to be displayed as a graph expressing an index value indicating a comparison with values ​​for the same age group based on axes corresponding to the multiple regions of interest.

8. A brain image analysis device as described in claim 1, wherein the plurality of structural units include the hippocampus and the whole brain volume, and the contrast calculation unit: i) generates data displaying the volume of the hippocampus as a first axis and the volume of the whole brain as a second axis perpendicular to the first axis; and ii) analyzes the degree of atrophy of the specified region of interest in the brain depending on which quadrant of the coordinates represented by the first axis and the second axis the calculated value of the brain volume of the subject belongs to.

9. A brain image analysis device as described in claim 1, further comprising a structural unit selection unit that selects a structural unit that is related to atrophy of the region of interest and changes less due to the influence of shooting conditions than other structural units based on the brain volume data stored in the database, and the identification unit identifies the structural unit selected by the structural unit selection unit.

10. A brain image analysis system that receives input of a brain tomographic image of a subject captured by a tomographic image scanner device and three-dimensionally analyzes the degree of atrophy of multiple regions of interest in the brain, comprising: a data server that stores a database of age-related change data including brain volume data of each of the structural units at each age under various imaging conditions for multiple subjects; and a brain image analysis server that analyzes the degree of atrophy of the multiple regions of interest, wherein the brain image analysis server includes: a segmentation unit that performs segmentation to divide the entire brain tomographic image into each structural unit; an identification unit that identifies a structural unit related to the region of interest from among the divided structural units; a volume calculation unit that calculates the brain volume of each of the identified structural units; and a comparison calculation unit that calculates an index value indicating a comparison of the volume of the structural unit of the subject to be evaluated with the value of the same age from the data of each structural unit at each age in the database, The contrast calculation unit analyzes the degree of atrophy of the specified region of interest in the brain by integrating and analyzing multiple structural units for each age in the database and calculating the index value.

11. A brain image analysis program for causing a computer having a storage device and a calculation device to execute processing as an analysis device that receives input of a brain tomographic image of an evaluation subject captured by a tomographic image scanner device and three-dimensionally analyzes the degree of atrophy of a specified region of interest in the brain, wherein the storage device has a database containing age-related change data including brain volume data of each of the structural units at each age under various imaging conditions, the calculation device performing a segmentation step of dividing the entire brain tomographic image into each structural unit, the calculation device identifying a structural unit related to the region of interest from among the divided structural units, the calculation device calculating the brain volume of each of the identified structural units, and the calculation device calculating an index value indicating a comparison of the volume of the structural unit of the evaluation subject with the value of the same age from the data of each of the structural units at each age in the database, A brain image analysis program, wherein the step of calculating the index value includes a step of integrating and analyzing a plurality of the structural units for each age in the database and calculating the index value, thereby analyzing the degree of atrophy of the specified region of interest in the brain.

Images