Cerebral cortex model-based simplicial network for diagnosing mental disorder

Abstract

A cerebral cortex model-based simplicial network for diagnosing mental disorder is disclosed. A cerebral cortex model-based method for diagnosing mental disorder performed by a mental disorder diagnosis system, according to one embodiment, may comprise the steps of: modeling a cerebral cortex mesh as a cerebral cortex simplicial complex; and performing message passing for diagnosing a mental disorder using relationship information of the modeled cerebral cortex simplicial complex through a message passing neural network.

Description

A Simplistic Network for Psychiatric Diagnosis Based on a Cerebral Cortex Model

[0001] The description below is about the diagnostic techniques for mental illness.

[0002]

[0003] Diagnosing mental illness using cerebral cortex models extracted from magnetic resonance imaging (MRI) data offers the advantage of leveraging geometric information about the cerebral cortex. Previous studies have treated the cerebral cortex mesh as a graph consisting of 40,962 nodes, using graph convolutional neural networks (GCNs) or spherical convolutional neural networks (SCNs). However, these two existing approaches have limitations in directly modeling the high-dimensional relationships within the graph structure. In cerebral cortex data, each vertex can be viewed as a cortical thickness corresponding to a three-dimensional position. When modeling these cortical features, the relationship between vertices in the graph structure is simply considered as a pair of thickness information, which does not sufficiently reflect the biological and geometric information of the cerebral cortex.

[0004]

[0005] The cerebral cortex mesh can be modeled as a cerebral cortex group complex, and message passing for diagnosing mental illness can be performed using the modeled cerebral cortex group complex through a message passing neural network.

[0006]

[0007] A mental illness diagnosis method based on a cerebral cortex model performed by a mental illness diagnosis system may include a step of modeling a cerebral cortex mesh as a cerebral cortex simplex complex; and a step of performing message passing for mental illness diagnosis using relationship information of the modeled cerebral cortex simplex complex through a message passing neural network.

[0008] The above modeling step may include a step of generating a matrix representing connectivity in a cerebral cortical group complex through an adjacency matrix representing connectivity between vertices in a cerebral cortical mesh generated through brain imaging processing software.

[0009] The above brain imaging processing software may use brain imaging data to generate a cerebral cortex mesh having a plurality of vertices in each hemisphere including a left hemisphere and a right hemisphere, and the plurality of vertices may have, as feature information, thickness information of the cerebral cortex that appears based on three-dimensional Euclidean distance data between the white matter and the gray matter.

[0010] In the above cerebral cortex mesh, a vertex is a 0-simplex in the cerebral cortex group complex, an edge is a 1-simplex, and a 2-simplex composed of multiple vertices can be defined based on the vertices and edges.

[0011] The step of performing the above may include defining a message passing layer of a cerebral cortex group complex using a matrix representing connectivity in the cerebral cortex group complex, and constructing a message passing neural network including message passing layers of a plurality of cerebral cortex group complexes defined above.

[0012] The step of performing the above may include a step of updating feature information for each vertex through information exchange using complex information of a simplex of the same dimension and complex information of adjacent different simplexes.

[0013] The step of performing the above may include a step of extracting feature information of each vertex through the defined message passing layer and global average pooling layer, and deriving a result prediction value through a classification layer for the extracted feature information.

[0014] A computer program stored in a computer-readable storage medium for executing a mental illness diagnosis method based on a cerebral cortex model performed by a mental illness diagnosis system, wherein the mental illness diagnosis method based on the cerebral cortex model may include a step of modeling a cerebral cortex mesh as a cerebral cortex simplex complex; and a step of performing message passing for mental illness diagnosis using relationship information of the modeled cerebral cortex simplex complex through a message passing neural network.

[0015] A mental illness diagnosis system may include a modeling unit that models a cerebral cortex mesh as a cerebral cortex simplex complex; and a message passing unit that performs message passing for mental illness diagnosis using relationship information of the modeled cerebral cortex simplex complex through a message passing neural network.

[0016] The above modeling unit can generate a matrix expressing connectivity in a cerebral cortex group complex through an adjacency matrix expressing connectivity between vertices in a cerebral cortex mesh generated through brain image processing software.

[0017] The above brain imaging processing software may use brain imaging data to generate a cerebral cortex mesh having a plurality of vertices in each hemisphere including a left hemisphere and a right hemisphere, and the plurality of vertices may have, as feature information, thickness information of the cerebral cortex that appears based on three-dimensional Euclidean distance data between the white matter and the gray matter.

[0018] In the above cerebral cortex mesh, a vertex is a 0-simplex in the cerebral cortex group complex, an edge is a 1-simplex, and a 2-simplex composed of multiple vertices can be defined based on the vertices and edges.

[0019] The above message passing unit can define a message passing layer of a cerebral cortex group complex using a matrix expressing connectivity in the cerebral cortex group complex, and construct a message passing neural network including message passing layers of a plurality of cerebral cortex group complexes defined above.

[0020] The above message passing unit can update feature information for each vertex through information exchange using complex information of the same simplex and complex information of different adjacent simplexes.

[0021] The above message passing unit can extract feature information of each vertex through the above-defined message passing layer and global average pooling layer, and derive a result prediction value through a classification layer for the extracted feature information.

[0022]

[0023] Compared to cerebral cortex models modeled using conventional graph convolutional networks or spherical convolutional neural networks, approaches utilizing complex groups can directly model more complex relationships between vertices. This not only enriches the feature set, but also facilitates learning broader information exchange through the connections between complexes of various dimensions.

[0024] Existing cortex-based methods utilize cortical thickness information at vertex locations to compute feature values, primarily focusing on limited inter-vertex interaction and local feature transfer. In contrast, high-dimensional message passing using group complexes can utilize a wider receptive field within a single layer than existing methods, enabling more effective information processing.

[0025]

[0026] FIG. 1 is a diagram illustrating the structure of a message passing neural network in one embodiment.

[0027] FIG. 2 is a diagram for explaining the detailed structure of a message passing neural network in one embodiment.

[0028] FIG. 3 is an example for explaining a sequential message transmission operation in one embodiment.

[0029] Figure 4 is an example for explaining a pooling operation in one embodiment.

[0030] FIG. 5 is a block diagram illustrating a mental illness diagnosis system in one embodiment.

[0031] FIG. 6 is a flowchart illustrating a method for diagnosing mental illness based on a cerebral cortex model in one embodiment.

[0032]

[0033] Hereinafter, the detailed description will be given with reference to the attached drawings.

[0034]

[0035] FIG. 1 is a diagram illustrating the structure of a message passing neural network in one embodiment.

[0036] A mental illness diagnosis system can model the cerebral cortex mesh (cerebral cortex model) as a cerebral cortical simplex complex. A cerebral cortical simplex complex is a type of high-dimensional graph that can be composed of 0-dimensional complexes, 1-dimensional complexes, and 2-dimensional complexes. For example, if one cerebral cortical simplex complex has three 0-dimensional complexes (nodes), three 1-dimensional (edge) complexes connecting the three 0-dimensional complexes, and one 2-dimensional complex surrounding the 0-dimensional complex and the 1-dimensional complex, this is defined by the formula {v0, v1, v2}, {{v0, v1}, {v1, v2}, {{v0, v1, v2}}. The feature values ​​of each 0-dimensional complex are calculated during the CIVET preprocessing process. Since the single-complex has combinatorial characteristics, when calculating the feature value of a one-dimensional complex, the feature values ​​of the constituent zero-dimensional complexes are linearly interpolated. For example, the feature value of {v0, v1} is calculated as a linear combination of the feature values ​​of {v0} and {v1}. Similarly, the feature value of the two-dimensional complex {v0, v1, v2} is calculated through linear interpolation using the feature values ​​of {v0}, {v1}, and {v2}.

[0037] A mental illness diagnosis system can calculate high-dimensional adjacency of a cerebral cortex mesh. When the cerebral cortex mesh is viewed as a cerebral cortical single-complex, each vertex can be viewed as a 0-simplex, each edge as a 1-simplex, and each triangle as a 2-simplex. Here, a node represents the cortical thickness of a specific region, an edge represents the cortical thickness between nodes, and a face (triangle) represents the cortical thickness of an area connected by three nodes. The mental illness diagnosis system can learn a message-passing neural network using the structure of the modeled cerebral cortical single-complex. At this time, the mental illness diagnosis system can build a message-passing neural network based on the simplex complex neural network. Simplicial complex neural network is a methodology that can directly model the relationship between three nodes and consider various types of adjacency, overcoming the shortcoming of not being able to directly model high-dimensional relationships in graphs as a higher-order concept of graphs.

[0038] A mental illness diagnosis system can consider multiple types of associations for each simplex. For example, associations that can be considered for an edge can be categorized as boundary, coboundary, upper Laplacian, and lower Laplacian. Boundary indicates connectivity with directly connected vertices. Coboundary indicates connectivity with directly connected triangles. High Laplacian indicates connectivity between edges that share the same triangle. Low Laplacian indicates connectivity between edges that share the same vertex.

[0039] A mental illness diagnosis system can utilize connectivity information for each vertex in a cerebral cortex mesh acquired from magnetic resonance imaging. Based on the correlations between each cerebral cortex complex, the system can pre-compute the boundary relation matrix and the co-boundary relation matrix. The system can compute the high and low Laplacians for all dimensions by multiplying the boundary relation matrix and the co-boundary relation matrix.

[0040] A mental illness diagnosis system can perform high-dimensional message passing. It can compute a high-dimensional association matrix to calculate feature information for each dimension. During this process, each complex in each dimension can update its feature values ​​by utilizing information from not only complexes in the same dimension but also adjacent complexes in other dimensions. The cerebral cortex is composed of 0-dimensional, 1-dimensional, and 2-dimensional complexes (simplexes), and message passing proceeds sequentially from 2-dimensional to 0-dimensional. Referring to Figure 3, an example illustrates the sequential message passing operation. From right to left, triangle-level message passing, edge-level message passing, and vertex-level message passing are performed. More specifically, when updating the feature values ​​of a 2-dimensional complex, adjacent 2-dimensional and 1-dimensional complexes can be considered. The feature value transfer from 1-dimensional to 2-dimensional can be calculated by multiplying the co-boundary relationship matrix of the 2-dimensional complex and the feature value matrix of the 1-dimensional complex. Feature transfer within a two-dimensional complex can be achieved by multiplying the lower Laplacian matrix by the two-dimensional feature matrix. During this process, weight matrices corresponding to each feature can be multiplied. Furthermore, feature calculations for a one-dimensional complex can consider four correlation matrices. The first involves multiplying the two-dimensional feature matrix calculated in the previous step by the one-dimensional boundary relation matrix, which represents information transfer from two dimensions to one. The second and third involve information transfer within the one-dimensional complex, which can be calculated by multiplying the higher and lower Laplacian matrices. Finally, information transfer from one dimension to two dimensions can be calculated using the one-dimensional co-boundary relation matrix. Here, weight matrices corresponding to the four features can also be multiplied. Finally, feature calculations for a zero-dimensional complex can be achieved by multiplying the new one-dimensional feature matrix by the boundary relation matrix.Information transfer within a 0-dimensional space can be computed using a high Laplacian matrix. This process also involves multiplying each feature value by a corresponding weight matrix.

[0041] Mental illness diagnosis systems can utilize hierarchical feature values ​​through pooling layers. In conventional spherical graphs, the feature dimension is typically extracted by extracting the maximum or average value among neighboring nodes in a 1-ring. Spherical convolutional neural networks, on the other hand, utilize an inverse expansion process that considers the segmentation characteristics of the icosahedron and restores the segmented shape to its original state. In the embodiment, the adoption of the inverse process demonstrates superior efficiency and performance.

[0042] FIG. 2 is a diagram for explaining the detailed structure of a message passing neural network in one embodiment.

[0043] Assessing brain atrophy is crucial in diagnosing dementia, a neurodegenerative disease. Cortical thickness data is a key piece of information used in this assessment. Psychiatric diagnosis systems can use a measurement method based on a mid-surface model to measure cortical thickness. Image preprocessing can be performed automatically in a Linux computer environment using software called CIVET. CIVET is a brain imaging software that automatically generates brain structural and surface models from MRI brain images. Through this process, the psychiatric diagnosis system can use the CLASP transformational modeling algorithm to generate mesh-based gray and white matter models with 40,962 vertices in each hemisphere. Each vertex on the mesh represents cortical thickness based on the three-dimensional Euclidean distance between white and gray matter. The psychiatric diagnosis system can generate a cortical mesh using the CIVET algorithm by receiving MRI brain images. At this time, each vertex on the cerebral cortex mesh has the cerebral cortex thickness value at that location as a feature value.

[0044] A mental illness diagnosis system can calculate a matrix representing four connectivity types on a cortical single-unit complex from a cortical mesh using an adjacency matrix representing connectivity between vertices in the cortical mesh. The adjacency matrix representing connectivity between vertices on the cortical mesh can be calculated from the cortical mesh. This process generates a matrix representing four connectivity types in the single-unit complex. Vertices in the cortical mesh correspond to 0-simplexes in the single-unit complex, and edges represent 1-simplexes. Based on this, a 2-simplex complex consisting of three vertices can be defined. The mental illness diagnosis system can use the vertex adjacency matrix to search for vertices represented by triangles to form a 2-simplex set. After forming each set for 0-simplexes, 1-simplexes, and 2-simplexes, the system can calculate a matrix representing connectivity between complexes whose dimensionality differs by 1. For a 1-simplex, the matrix value corresponding to the 0-simplex that constitutes the 1-simplex is defined as 1, otherwise 0. The matrix representing the adjacency of the 0-simplex and the 1-simplex thus generated means the co-boundary relationship of the 0-simplex and the boundary relationship of the 1-simplex. For a 2-simplex, the matrix value corresponding to the 1-simplex that constitutes the 2-simplex is defined as 1, otherwise 0. The matrix representing the adjacency of the 1-simplex and the 2-simplex thus generated means the co-boundary relationship of the 1-simplex and the boundary relationship of the 2-simplex. The Laplacian matrix representing the connectivity between complexes of the same dimension can also be calculated. The Laplacian matrix for a 0-simplex represents the connectivity between 0-simplexes, and the matrix value is defined as 1 if there is a shared 1-simplex and 0 otherwise. This is the upper Laplacian matrix of the 0-simplex.The Laplacian matrix of the 1-simplex dimension represents the connectivity between 1-simplexes, and is defined as 1 if there is a shared 0- or 2-simplex, and 0 otherwise. These are the lower Laplacian and upper Laplacian matrices of the 1-simplex, respectively. The lower and upper Laplacians can be combined to compute the Laplacian representing the connectivity between 1-simplexes. Similarly, the lower Laplacian matrix of the 2-simplex dimension can also be computed.

[0045] A mental illness diagnosis system can define a high-dimensional message-passing layer using a matrix representing the four-way connectivity between vertices on the cerebral cortical network complex. Since cortical thickness values ​​are derived from the 0-simplex, the feature values ​​of the 1-simplex and 2-simplex can be calculated through linear interpolation of the feature values ​​of the 0-simplexes they comprise.

[0046] A mental illness diagnosis system can define information exchange between complexes using a feature matrix and a connectivity matrix. First, the system can compute the feature values ​​of the highest-dimensional 2-simplex. Here, two paths of information exchange are considered. First, the system can compute new feature values ​​by multiplying the lower Laplacian matrix between 2-simplexes and the feature matrix of the 2-simplex. Second, the system can use a boundary matrix to include the feature values ​​of the 2-simplex and its adjacent 1-simplex. Afterwards, the system can obtain the final 2-simplex feature values ​​by summing the two results.

[0047] Next, the mental illness diagnosis system can define the feature value information exchange of 1-simplexes. Here, we consider four paths of information exchange. First, the mental illness diagnosis system can multiply the lower and upper Laplacians of 1-simplexes by the feature value of the 1-simplex and add the results. Second, the mental illness diagnosis system can use a boundary matrix to include the feature values ​​of the 1-simplex and the adjacent 0-simplexes. Third, the mental illness diagnosis system can use a co-boundary matrix to include the feature values ​​of the 1-simplex and the adjacent 2-simplexes. Finally, the mental illness diagnosis system can obtain the final 1-simplex feature value by summing the three results.

[0048] A mental illness diagnosis system can define a feature information exchange process for the lowest-dimensional 0-simplex. This information exchange occurs through two pathways. First, the system can multiply the lower Laplacian matrix between 0-simplexes by the feature values ​​of the 0-simplexes. This process considers the direct connection between 0-simplexes. Second, the system can use a co-boundary matrix to include the feature values ​​of the 0-simplex and its adjacent 1-simplexes. This matrix reflects the relationship between the 0-simplex and the higher-dimensional complexes surrounding it. The system can then sum the two calculation results to derive a new feature value for the 0-simplex. Through this process, the system can calculate a new feature value for each vertex in the cerebral cortical group complex.

[0049] The mental illness diagnosis system can define a high-dimensional message passing layer of the cerebral cortex group complex and then build a group complex high-dimensional message passing neural network that includes the message passing layer. The message passing neural network can process a cerebral cortex mesh consisting of 40,962 vertices. The message passing neural network can utilize a convolutional neural network on spherical data to generate richer feature values ​​for each vertex.

[0050] Referring to Figure 4, an example illustrates a pooling layer. The mental illness diagnosis system can extend the feature dimension of each vertex to 512 by repeatedly applying a high-dimensional message-passing layer to a defined cerebral cortex group complex four times. At this time, representative feature values ​​can be extracted for each hemisphere through a global average pooling layer. These extracted features are combined and ultimately predicted through a classifier layer.

[0051] Because changes in the cerebral cortex appear in the early stages of Alzheimer's disease, analyzing cerebral cortex data can facilitate early detection. In one embodiment, a deep learning model utilizing cerebral cortex data can provide objective and quantitative assessments.

[0052] FIG. 5 is a block diagram for explaining a mental illness diagnosis system in one embodiment, and FIG. 6 is a flowchart for explaining a mental illness diagnosis method based on a cerebral cortex model in one embodiment.

[0053] The processor of the mental illness diagnosis system (100) may include a modeling unit (510) and a message passing unit (520). The processor may be expressions of different functions performed by the processor according to control commands provided by program codes stored in the mental illness diagnosis system of the processor. The processor and components of the processor may control the mental illness diagnosis system that performs steps (610 to 620) included in the mental illness diagnosis method based on the cerebral cortex model of FIG. 6. At this time, the processor and components of the processor may be implemented to execute instructions according to the code of the operating system included in the memory and the code of at least one program.

[0054] The processor can load program code stored in a program file for a mental illness diagnosis method based on a cerebral cortex model into memory. For example, when the program is executed in the mental illness diagnosis system, the processor can control the simulation system to load the program code from the program file into memory under the control of the operating system. At this time, the processor can execute instructions of corresponding portions of the program code loaded into memory in each of the modeling unit (510) and the message passing unit (520), thereby executing subsequent steps (610 to 620), which can be different functional representations of the processor.

[0055] In step (610), the modeling unit (510) can model the cerebral cortex mesh as a cerebral cortex simplex complex. The modeling unit (510) can generate a matrix expressing connectivity in the cerebral cortex simplex complex through an adjacency matrix expressing connectivity between vertices in the cerebral cortex mesh generated by brain imaging processing software. At this time, the brain imaging processing software can generate a cerebral cortex mesh having multiple vertices in each hemisphere, including a sphere or the left hemisphere and the right hemisphere, using brain imaging data.

[0056] In step (620), the message passing unit (520) can perform message passing for diagnosing mental illness using relationship information of a cerebral cortex group complex modeled through a message passing neural network. The message passing unit (520) can define a message passing layer of a cerebral cortex group complex using a matrix expressing connectivity in the cerebral cortex group complex, and build a message passing neural network including message passing layers of a plurality of defined cerebral cortex group complexes. The message passing unit (520) can update feature information for each vertex through information exchange using complex information of a simplex of the same dimension and complex information of different adjacent simplexes. The message passing unit (520) can extract feature information for each vertex through a defined message passing layer and a global average pooling layer, and derive a result prediction value for the extracted feature information through a classification layer.

[0057]

[0058] The devices described above may be implemented as hardware components, software components, and / or a combination of hardware components and software components. For example, the devices and components described in the embodiments may be implemented using one or more general-purpose computers or special-purpose computers, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing instructions and responding to them. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing device is sometimes described as being used alone; however, one of ordinary skill in the art will recognize that the processing device may include multiple processing elements and / or multiple types of processing elements. For example, a processing unit may include multiple processors, or a processor and a controller. Other processing configurations, such as parallel processors, are also possible.

[0059] Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing device to perform a desired operation or may independently or collectively command the processing device. The software and / or data may be embodied in any type of machine, component, physical device, virtual equipment, computer storage medium, or device for interpretation by the processing device or for providing instructions or data to the processing device. The software may be distributed over networked computer systems and stored or executed in a distributed manner. The software and data may be stored on one or more computer-readable recording media.

[0060] The method according to the embodiment may be implemented in the form of program commands that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program commands, data files, data structures, etc., alone or in combination. The program commands recorded on the medium may be those specially designed and configured for the embodiment or may be those known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program commands such as ROMs, RAMs, and flash memories. Examples of program commands include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc.

[0061] Although the embodiments described above have been described by way of limited examples and drawings, those skilled in the art will appreciate that various modifications and variations can be made based on the above teachings. For example, appropriate results can still be achieved even if the described techniques are performed in a different order than described, and / or components of the described systems, structures, devices, circuits, etc. are combined or combined in a different manner than described, or are replaced or substituted with other components or equivalents.

[0062] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims described below.

Claims

1. In a mental illness diagnosis method based on a cerebral cortex model performed by a mental illness diagnosis system, A step of modeling the cerebral cortex mesh as a cerebral cortex simplex complex; and A step of performing message passing for diagnosing mental illness by using the relationship information of the modeled cerebral cortex group complex through a message passing neural network. A method for diagnosing mental illness based on a cerebral cortex model including:

2. In paragraph 1, The above modeling steps are: A step of generating a matrix representing connectivity in a cerebral cortex complex through an adjacency matrix representing connectivity between vertices in a cerebral cortex mesh generated by brain imaging processing software. A method for diagnosing mental illness based on a cerebral cortex model including:

3. In paragraph 2, The above brain imaging processing software uses brain imaging data to generate a cerebral cortex mesh having multiple vertices in each hemisphere, including a left hemisphere and a right hemisphere, and The above plurality of vertices have as feature information the thickness information of the cerebral cortex that appears based on the three-dimensional Euclidean distance data between the white matter and the gray matter. A method for diagnosing mental illness based on a cerebral cortex model characterized by:

4. In paragraph 2, A mental illness diagnosis method based on a cerebral cortex model, characterized in that in the above cerebral cortex mesh, a vertex is a 0-simplex in a cerebral cortex group complex, an edge is a 1-simplex, and a 2-simplex composed of a plurality of vertices is defined based on the vertices and edges.

5. In paragraph 1, The steps performed above are: A step of defining a message passing layer of a cerebral cortex group complex using a matrix representing connectivity in the cerebral cortex group complex, and constructing a message passing neural network including message passing layers of a plurality of cerebral cortex group complexes defined above. A method for diagnosing mental illness based on a cerebral cortex model including:

6. In paragraph 5, The steps performed above are: A step of updating feature information for each vertex by exchanging information using complex information of the same dimension simplex and complex information of different adjacent simplexes. A method for diagnosing mental illness based on a cerebral cortex model including:

7. In paragraph 5, The steps performed above are: A step of extracting feature information of each vertex through the above-defined message passing layer and global average pooling layer, and deriving a predicted result value through a classification layer for the extracted feature information. A method for diagnosing mental illness based on a cerebral cortex model including:

8. A computer program stored in a computer-readable storage medium for executing a mental illness diagnosis method based on a cerebral cortex model performed by a mental illness diagnosis system, The above cerebral cortex model-based mental illness diagnosis method is, A step of modeling the cerebral cortex mesh as a cerebral cortex simplex complex; and A step of performing message passing for diagnosing mental illness by using the relationship information of the modeled cerebral cortex group complex through a message passing neural network. A computer program stored on a computer-readable storage medium that executes the program.

9. In the mental illness diagnosis system, A modeling unit that models the cerebral cortex mesh as a cerebral cortex simplex complex; and A message passing unit that performs message passing for diagnosing mental illness by using the relationship information of the modeled cerebral cortex group complex through a message passing neural network. A mental illness diagnostic system that includes:

10. In paragraph 9, The above modeling section Generating a matrix representing connectivity in the cerebral cortex group complex through an adjacency matrix representing connectivity between vertices in a cerebral cortex mesh generated by brain imaging processing software. A mental illness diagnosis system characterized by:

11. In paragraph 10, The above brain imaging processing software uses brain imaging data to generate a cerebral cortex mesh having multiple vertices in each hemisphere, including a left hemisphere and a right hemisphere, and The above plurality of vertices have as feature information the thickness information of the cerebral cortex that appears based on the three-dimensional Euclidean distance data between the white matter and the gray matter. A mental illness diagnosis system characterized by:

12. In paragraph 10, A mental illness diagnosis system characterized in that in the above cerebral cortex mesh, a vertex is a 0-simplex in a cerebral cortex group complex, an edge is a 1-simplex, and a 2-simplex composed of a plurality of vertices is defined based on the vertices and edges.

13. In paragraph 9, The above message passing part is, A message passing layer of a cerebral cortex group complex is defined using a matrix representing connectivity in the cerebral cortex group complex, and a message passing neural network is constructed including message passing layers of a plurality of cerebral cortex group complexes defined above. A mental illness diagnosis system characterized by:

14. In paragraph 13, The above message passing part is, Update feature information for each vertex by exchanging information using complex information of the same simplex and complex information of different adjacent simplexes. A mental illness diagnosis system characterized by:

15. In paragraph 13, The above message passing part is, Extract feature information of each vertex through the message passing layer and global average pooling layer defined above, and derive a result prediction value through the classification layer for the extracted feature information. A mental illness diagnosis system characterized by:

Images