Brain function gradient analysis method, apparatus, medium, and program product

Abstract

The application belongs to the field of brain image data processing, and particularly relates to a brain function gradient analysis method, equipment, medium and program product. The brain function gradient analysis method comprises the following steps: obtaining a gradient space of a first function network and a gradient space of a second function network, the first function network and the second function network being different brain same function networks or same brain different function networks, and the function network being a set comprising at least two brain regions functionally associated with each other; and calculating a bulldozer distance between the gradient space of the first function network and the gradient space of the second function network to obtain a gradient distance. The bulldozer distance is used to capture spatial features such as shape, range and dispersion, so that the image data is fully mined. The improved brain function gradient analysis method based on the bulldozer distance can more effectively identify post-traumatic stress disorder.

Description

Technical Field

[0001] This application relates to the field of brain imaging data processing, and more specifically, to a brain functional gradient analysis method, device, medium, and program product. Background Technology

[0002] The concept of the human brain connectome provides a completely new perspective for understanding the brain's nervous system and offers an effective tool for studying the brain as a complete system. To fully utilize data from neuroimaging, researchers have been dedicated to exploring different data analysis methods, including static functional connectivity (sFC) and dynamic functional connectivity (dFC) for functional MRI (fMRI), and white matter connectivity for diffusion-weighted MRI. Based on these analytical methods, they have further explored the brain pathological manifestations of neuropsychiatric disorders (e.g., Alzheimer's disease, epilepsy, autism, depression, bipolar disorder, post-traumatic stress disorder, etc.).

[0003] While long-standing research methods linking measurements of neural processing (e.g., from functional magnetic resonance imaging, fMRI) to cognition have focused on identifying discrete brain regions and modules and their specific functional roles, recent conceptual and methodological developments have improved data analysis methods to allow mapping large-scale brain functions to low-dimensional manifold representations, also known as gradients. The research paper "Situating the default-mode network along a principal gradient of macroscale cortical organization" (DOI: 10.1073 / pnas.1608282113) discloses a typical gradient identification workflow, and gradient identification and gradient-based analysis methods are widely used in neuroscience research.

[0004] Post-traumatic stress disorder (PTSD) is a mental disorder caused by an exceptionally intense traumatic event (such as war, violent attack, serious accident, or natural disaster). Its core symptoms include re-experiencing the trauma, avoidance, negative cognitive and emotional changes, and increased vigilance. However, not all individuals who experience similar traumatic events develop PTSD. Some individuals experience trauma but do not exhibit significant psychopathological symptoms, or experience only brief symptoms followed by rapid recovery, demonstrating strong psychological resilience. This group is defined in studies as the Trauma-Exposed Control (TEC). Currently, in clinical practice and scientific research, the diagnosis and differentiation of PTSD mainly rely on interview-based clinical scales (such as CAPS-5) and patient self-report scales (such as PCL-5). These methods are highly dependent on patient subjective reports and clinician experience, and thus involve significant subjectivity. The key challenge is that while PTSD patients and TEC individuals share a common experience—exposure to a traumatic event—their pathophysiological manifestations and mechanisms are distinctly different. This "homologous but different fruit" characteristic makes it extremely difficult to differentiate based solely on clinical manifestations and subjective symptoms. Summary of the Invention

[0005] A key step in current gradient-based analysis methods is to estimate spatial separation by calculating the Euclidean distance between the centroids of the brain's functional networks. However, this approach has a drawback: the Euclidean distance calculation method simplifies the entire network into a single representative point, thus sacrificing spatial features such as the network's shape, extent, and dispersion. This is inconsistent with the goal of researchers to fully extract information from image data.

[0006] In view of the above problems, this application provides a brain functional gradient analysis method that uses bulldozer distance instead of Euclidean distance, which is widely used in the prior art, to estimate spatial separation, thereby achieving full mining of image data.

[0007] In a first aspect, the present invention discloses a brain functional gradient analysis method, comprising: obtaining the gradient space of a first functional network and the gradient space of a second functional network, wherein the first functional network and the second functional network are the same functional networks of different brains or different functional networks of the same brain, and the functional network is a set including at least two functionally related brain regions; calculating the bulldozer distance between the gradient space of the first functional network and the gradient space of the second functional network to obtain the gradient spacing.

[0008] Furthermore, when the first functional network and the second functional network contain the same number of brain regions, the gradient spacing is obtained by calculating the bulldozer distance between the gradient spaces of the first functional network and the second functional network; the bulldozer distance is expressed as:

[0009]

[0010] in, Indicates the distance to the bulldozer. This represents the gradient space of the first functional network. This represents the gradient space of the second functional network. and Since they contain the same number of brain regions, min() represents finding the minimum value. Indicates from Convert to A collection of effective transfer plans, Represents any one of the valid transfer plans in the set of valid transfer plans. yes The inner product of the matrix M and M, where M is and The transition cost matrix between them.

[0011] Furthermore, when the first functional network and the second functional network contain different numbers of brain regions, a partial bulldozer distance is calculated between the gradient space of the first functional network and the gradient space of the second functional network to obtain the gradient spacing, wherein the partial bulldozer distance is expressed as:

[0012]

[0013] in, This represents a partial bulldozer distance between the gradient space of the first functional network and the gradient space of the second functional network. This represents the gradient space of the first functional network. This represents the gradient space of the second functional network. and The number of brain regions included differs. M represents a feasible portion of the overall transfer plan. and The transition cost matrix between them yes The inner product of the matrix M and M, min() means finding the minimum value.

[0014] Furthermore, the transfer cost matrix represents the transfer cost required to convert the gradient space of the first functional network into the gradient space of the second functional network. The element in the i-th row and j-th column of the transfer cost matrix represents the transfer cost of converting the gradient coordinates of the i-th cortical partition in the gradient space of the first functional network into the gradient coordinates of the j-th cortical partition in the gradient space of the second functional network. The transfer cost is represented by the L-norm. The element in the i-th row and j-th column of the transfer plan represents the transfer quality of converting the gradient coordinates of the i-th cortical partition in the gradient space of the first functional network into the gradient coordinates of the j-th cortical partition in the gradient space of the second functional network.

[0015] Furthermore, the method for obtaining the gradient space of the first functional network includes: acquiring functional magnetic resonance imaging of the first subject; performing functional connectivity analysis based on the functional magnetic resonance imaging to obtain a first brain functional connectivity matrix; performing dimensionality reduction and gradient calculation on the first brain functional connectivity matrix to obtain a first low-dimensional gradient space; and extracting the coordinates on the set gradient of the first functional network in the first low-dimensional gradient space to obtain the gradient space of the first functional network.

[0016] Furthermore, the set gradient is expressed as: , It is a set An element in the power set, where D is the number of gradients contained in the low-dimensional gradient space.

[0017] Secondly, this invention discloses a method for predicting post-traumatic stress disorder, the method comprising:

[0018] Acquire functional magnetic resonance imaging (fMRI) images of the subject; perform functional connectivity analysis based on the fMRI images to obtain a brain functional connectivity matrix; perform dimensionality reduction and gradient calculation on the brain functional connectivity matrix to obtain a low-dimensional gradient space, which includes a first gradient, a second gradient, and a third gradient; extract gradient spaces on a set gradient of at least two functional networks based on the low-dimensional gradient space to obtain the gradient spaces of the first functional network and the second functional network, where the set gradient is the third gradient; calculate the gradient interval based on the gradient spaces of the first and second functional networks according to the brain functional gradient analysis method described above; input the gradient interval into a classifier to obtain the result of whether the subject suffers from post-traumatic stress disorder.

[0019] Furthermore, the gradient spacing includes one or more of the following: gradient spacing between the default mode network and the visual network; gradient spacing between the default mode network and the salience network; gradient spacing between the default mode network and the dorsal attention network; gradient spacing between the default mode network and the ventral attention network; gradient spacing between the default mode network and the sensorimotor network; gradient spacing between the default mode network and the edge network; gradient spacing between the visual network and the salience network; gradient spacing between the visual network and the dorsal attention network; gradient spacing between the visual network and the ventral attention network; gradient spacing between the visual network and the sensorimotor network; gradient spacing between the visual network and the edge network; gradient spacing between the salience network and the dorsal attention network; gradient spacing between the salience network and the ventral attention network; gradient spacing between the salience network and the sensorimotor network; gradient spacing between the salience network and the edge network; gradient spacing between the dorsal attention network and the ventral attention network; gradient spacing between the dorsal attention network and the sensorimotor network; gradient spacing between the dorsal attention network and the edge network; gradient spacing between the ventral attention network and the sensorimotor network; gradient spacing between the ventral attention network and the edge network; and gradient spacing between the sensorimotor network and the edge network.

[0020] Thirdly, this invention discloses a brain functional gradient analysis system, comprising:

[0021] First acquisition module: used to acquire the gradient space of the first functional network and the gradient space of the second functional network, wherein the first functional network and the second functional network are the same functional networks in different brains or different functional networks in the same brain, and the functional network is a set including at least two functionally related brain regions; First gradient distance calculation module: used to calculate the bulldozer distance between the gradient space of the first functional network and the gradient space of the second functional network to obtain the gradient distance.

[0022] Fourthly, this invention discloses a post-traumatic stress disorder prediction system, comprising:

[0023] The second acquisition module is used to acquire functional magnetic resonance imaging (fMRI) images of the subject; the brain functional connectivity matrix extraction module is used to perform functional connectivity analysis based on the fMRI images to obtain a brain functional connectivity matrix; and the low-dimensional gradient space extraction module is used to perform dimensionality reduction and gradient calculation on the brain functional connectivity matrix to obtain a low-dimensional gradient space, wherein the low-dimensional gradient space includes a first gradient, a second gradient, and a third gradient; the functional network gradient space gradient module is used to extract gradient spaces on a set gradient of at least two functional networks based on the low-dimensional gradient space to obtain the gradient space of the first functional network and the gradient space of the second functional network, wherein the set gradient is the third gradient; the gradient spacing calculation module is used to calculate the gradient spacing based on the gradient spaces of the first functional network and the gradient spaces of the second functional network according to the brain functional gradient analysis method described above; and the prediction module is used to input the gradient spacing into a classifier to obtain the result of whether the subject suffers from post-traumatic stress disorder.

[0024] Fifthly, the present invention discloses a computer device, the device comprising: a memory and a processor; the memory being used to store program instructions; the processor being used to invoke the program instructions, which, when executed, are used to perform the steps of the method described above.

[0025] In a sixth aspect, the present invention discloses a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0026] In a seventh aspect, the present invention discloses a computer program product, comprising a computer program that, when executed by a processor, implements the steps of the above-described method.

[0027] This application has the following beneficial effects:

[0028] (1) This application uses bulldozer distance instead of Euclidean distance, which is widely used in the prior art, to estimate spatial separation, captures spatial features such as shape, range and dispersion, and realizes full mining of image data;

[0029] (2) This application uses the improved brain functional gradient analysis method based on bulldozer distance to predict post-traumatic stress disorder (PTSD), which can more effectively identify PTSD. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the method flow provided in the first aspect of the embodiments of this application.

[0032] Figure 2 This is a schematic diagram of the program product provided in the second aspect of the embodiments of this application.

[0033] Figure 3 This is a schematic diagram of the architecture of an exemplary computing device provided in the embodiments of this application.

[0034] Figure 4 This is a schematic diagram of the main steps of gradient analysis in an embodiment of this application: where A is a schematic diagram of the data representation of the acquired fMRI image provided in an embodiment of this application; and where B is a schematic diagram of a gradient extraction process provided in an embodiment of this application.

[0035] Figure 5 This is a schematic diagram of the original probability and target probability distribution for bulldozer distance calculation provided in an embodiment of this application.

[0036] Figure 6 This is a schematic diagram of a transfer plan for bulldozing distance calculation provided in an embodiment of this application.

[0037] Figure 7 This application provides an embodiment of the transfer mass for bulldozing distance calculation. A schematic diagram of the matrix representation of the transfer cost D.

[0038] Figure 8 This is a schematic diagram of the gradient space of the HC, TEC, PTSD group containing different gradient combinations, represented on a coordinate system and on a brain network, according to an embodiment of this application: where a illustrates the gradient space of the HC, TEC, PTSD group containing three gradients and two gradients combined in pairs, represented on a coordinate system; where b illustrates the gradient space of the HC, TEC, PTSD group containing one gradient, represented on the cerebral cortex and on various brain networks.

[0039] Figure 9 This is a schematic diagram comparing the performance of models that identify PTSD and TEC based on network spacing, provided in an embodiment of this application. Detailed Implementation

[0040] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0041] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as S101, S102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0042] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0043] Figure 1 This is a schematic flowchart of a method for assessing brain functional gradient analysis provided in an embodiment of this application. Specifically, the method includes the following steps:

[0044] S101: Obtain the gradient space of the first functional network and the gradient space of the second functional network, wherein the first functional network and the second functional network are the same functional networks in different brains or different functional networks in the same brain, and the functional network is a set including at least two functionally related brain regions.

[0045] According to embodiments of this application, a functional network refers to a network in the brain consisting of at least two interconnected brain regions or nodes, wherein the interconnection is generally determined based on the temporal correlation, statistical dependence, or functional connectivity of brain region signals. Methods for defining brain regions or nodes include: 1) using each voxel / vertex as a node; 2) using predefined brain region atlases (such as Schaefer atlas 200 / 400 / 1000 areas, Glasser 360, AAL, etc.). All brain regions can be considered as a single functional network, i.e., a whole-brain functional network.

[0046] According to embodiments of this application, the whole-brain functional network can be divided into seven Yeo functional networks based on their different functions: ① Default Mode Network (DMN), ② Visual Network (VIS), ③ Salience Network (SN), ④ Dorsal Attention Network (DAN), ⑤ Ventral Attention Network (VAN), ⑥ Sensorimotor Network (SMN), and ⑦ Limbic Network (LIM).

[0047] According to embodiments of this application, in addition to the seven-function network, other functional network partitioning methods can also be used, such as Yeo's 17-function network, ICA-based network partitioning, etc.

[0048] According to an embodiment of this application, the gradient space of a functional network refers to the low-dimensional gradient space of the whole brain obtained after performing gradient analysis on the whole brain functional connectivity (FC) matrix, and then extracting the coordinates on a specific gradient of a specific functional network to obtain the gradient coordinates of the specific functional network (specific gradient).

[0049] According to an embodiment of this application, after performing gradient analysis on FC, the low-dimensional gradient space is represented as follows: Where GS (Gradient Space) represents the low-dimensional gradient space, This represents the low-dimensional gradient vector of the i-th brain region. N represents the number of nodes based on the whole brain or the number of brain regions obtained by dividing the brain based on a brain template. ,in, Let represent the coordinates of the j-th gradient of the low-dimensional gradient vector of the i-th brain region, which is a real value. G represents the number of gradients contained in the low-dimensional gradient space. Assuming the first functional network includes brain regions 1, 7, 47, 56, 90, and 100, and let the gradient be the second gradient, then the gradient space of the first functional network is represented as: Assuming the second functional network includes brain regions 39, 45, 55, 77, and 79, and its gradient is defined as the second gradient, then the gradient space representation of the defined gradient of the second functional network is as follows: Assuming the functional network includes brain regions 9, 37, 49, and 51, and the gradients are designated as the second and third gradients, then the gradient space representation of the designated gradients of this functional network is as follows: .

[0050] The main steps of gradient analysis are as follows: Figure 4 As shown in A and B: Step 1: Collect fMRI images of the subjects, and obtain time-series data of each node by defining brain nodes based on the data ( Figure 4 Step A); Step 2: Construct the Functional Connectivity (FC) matrix; matrix elements , indicating the strength of the functional connection between node i and node j ( Figure 4 Step 3: Construct the affinity matrix / association matrix, based on cosine distance: Where Aff represents the affinity matrix, This represents the calculation of the cosine distance between the connection patterns of various brain regions in the central nervous system (FC). Figure 4 Step B); Step 4: Dimensionality reduction and gradient calculation. The goal of dimensionality reduction and gradient calculation is to find a set of low-dimensional embedding coordinates, i.e., the gradient ( Figure 4 The B in the equation aims to make the distance between nodes in the low-dimensional space as close as possible to the distance defined by the original high-dimensional distance matrix Aff. The mainstream algorithm for dimensionality reduction and gradient calculation is Diffusion Map Embedding (DMAP). The core idea of ​​this algorithm is to simulate a random walk process and capture the geometric structure of the diffusion of functional connection patterns between nodes. It focuses more on local connection patterns (short-distance connections) and is relatively insensitive to the absolute value of global connection strength, which helps to reveal continuous changes in brain functional organization. The coordinates of each node in the low-dimensional gradient space (usually the first 2-3 dimensions) are obtained through dimensionality reduction and gradient calculation. , This represents the coordinates of the i-th node in gradient 1 (or the first gradient); This represents the coordinates of the i-th node in gradient 2. Step 5: Gradient Alignment (for multi-subject or group analyses): When analyzing multiple subjects, each subject calculates their own set of gradient coordinates. The gradient direction (positive or negative) and order (which is G1 and which is G2) are arbitrary between individuals, requiring alignment for group-level analysis or inter-group comparisons. Common methods for gradient alignment include: group mean template method and generalized Protodyakonov analysis. Generalized Protodyakonov analysis simultaneously optimizes the alignment of all individual gradients with a common template. After gradient alignment, the coordinates of all subjects are obtained in a low-dimensional gradient space defined with the same direction and order.

[0051] According to the embodiments of this application, the low-dimensional gradient space is obtained by calculating the spatial correlation of standard template gradients (G1_template, G2_template, G3_template, etc.). The standard gradient templates are published and maintained by the research team that disclosed the typical gradient recognition workflow in the background technology paper "Situating the default-mode network along a principal gradient of macroscale cortical organization" (DOI: 10.1073 / pnas.1608282113).

[0052] In one embodiment of this application, the study of collecting fMRI images of PTSD subjects and performing gradient analysis includes the following main research steps and contents. Step 1: Data collection and preprocessing. (1.1) Subject collection and grouping: The research team recruited 70 typhoon-exposed subjects from typhoon-affected areas. Initial screening was conducted using the civilian version of the Post-Traumatic Stress Disorder Checklist (PCL). Subjects with scores exceeding 35 points were further diagnosed using a semi-structured clinical interview (DSM-IV Axis I Disorders Pattern Clinical Examination, SCID-IV). According to the DSM-IV diagnostic criteria, 36 subjects were ultimately diagnosed with PTSD, while the remaining 34 (trauma exposure control group, TEC) did not reach the diagnostic threshold. The severity and progression of symptoms in diagnosed patients were assessed using the PTSD Scale for Clinicians (CAPS). This structured clinical interview quantified the frequency and intensity of 17 core PTSD symptoms using a 0-4 point behavioral anchoring scoring system. SCID-IV was also used to assess comorbid mental disorders. In addition, the study recruited 32 healthy controls (HCs) from Haikou, a neighboring city of Wenchang, who did not meet the A1 diagnostic criteria for PTSD, through public advertising. All participants completed the Self-Rating Depression Scale (SDS) and the Self-Rating Anxiety Scale (SAS) to assess their emotional symptoms. All clinical assessments were conducted by psychiatrists at the Second Xiangya Hospital of Central South University between November 2014 and January 2015. Exclusion criteria included: age <18 years or >65 years; serious neurological or physical illness; history of head trauma or loss of consciousness; left-handedness. Other exclusion factors included current or past mental disorders (excluding depression / anxiety), use of psychotropic drugs, alcohol / substance abuse, and contraindications to MRI (such as pregnancy, claustrophobia, presence of ferromagnetic implants, etc.). Three women in the PTSD group were excluded due to incomplete imaging data, and six others were excluded for the following reasons: cerebral infarction (1 woman), denture artifact (1 woman / 1 man), pregnancy (1 woman), and excessive head movement (1 woman / 1 man). Two males in the HC group were excluded due to cerebral infarction, and one female in the TEC group was excluded due to excessive head movement. The final sample included 27 PTSD patients, 33 TEC patients, and 30 HC patients. (1.2) Resting-state functional magnetic resonance imaging (fMRI) acquisition: Data acquisition was completed at Hainan General Hospital using a Siemens 3.0 Tesla Skyra MRI scanner (equipped with a 32-channel standard head coil). The data acquisition process was a routine procedure and will not be described in detail here. (1.3) Preprocessing process: Preprocessing was completed using the fMRIPrep 20.2.1 toolkit based on Nipype 1.5.1 (integrating FreeSurfer, FSL, ANTs, and AFNI tools). The preprocessing process was a routine procedure and will not be described in detail here.Step 2: Constructing Cortical Functional Gradients (i.e., dimensionality reduction and gradient calculation to obtain a low-dimensional gradient space): The BrainSpace toolbox was used to calculate the cortical gradient (i.e., gradient space) of functional connectivity. Based on the 7-network partitioning framework proposed by Yeo et al., the Schaefer atlas (Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic Functional Connectivity MRI, DOI: 10.1093 / cercor / bhx179) was used to extract the resting-state time series of each subject from 1000 cortical partitions. This atlas provides spatially continuous partitions corresponding to classic large-scale networks, achieving a fine mapping of cortical functional architecture while maintaining network-level interpretability. For each subject, a functional connectivity (FC) matrix was constructed by calculating the Pearson correlation coefficients between all cortical partition pairs, and Fisher's r-to-z transformation was applied to improve data normality and facilitate cross-subject comparisons. To highlight the most informative connections, the matrix is ​​thresholded, retaining the top 10% of strength values. Then, the cosine similarity of the interval connection patterns is calculated to generate an affinity matrix. This affinity matrix is ​​used as input to a diffusion map embedding (a nonlinear manifold learning technique) to reduce the high-dimensional connectivity structure to a low-dimensional representation, capturing the principal axes of cortical organization. In this process: the α parameter is set to 0.5 (α is a gradient algorithm parameter, often used to mitigate the non-uniform sampling effect of the cortical manifold); the diffusion time is set to 0 to enhance the fine distinction between cortical regions. Through the above steps, the group-averaged functional connectivity matrices for the HC, TEC, and PTSD groups are generated respectively, and the group-level gradient templates are derived accordingly, resulting in the group-level templates (or group low-dimensional gradient spaces) for the three groups, denoted as follows. , , Individual gradients were aligned with the corresponding group templates after being rotated using Procrustes. Subsequent analyses in this study focused on the first three major gradient components (which explained 52.37% of the total variance in cortical functional connectivity).

[0053] According to embodiments of this application, in order to improve the comparability of gradient dimensions between groups, the gradient of each subject is standardized according to the explained variance of its corresponding reference gradient—these reference gradients are derived from group-level templates after individual gradient alignment.

[0054] According to embodiments of this application, the calculated gradient network (i.e., individual low-dimensional gradient space) of subjects in the PTSD, TEC, or HC groups is obtained. Reference network (i.e., the low-dimensional gradient space of the reference network) Data dimensions are Where n represents the number of cortical parcels, and dim represents the dimension of the selected principal gradient. In this study, n=1000 and n=3 were used.

[0055] The gradient space of a functional network over a given gradient can be viewed as a discrete probability distribution. Each network is represented in the gradient space as a set of gradient coordinates of a partition. Let the gradient space of a functional network of an individual be represented as... The gradient space of the reference functional network is represented as: ,in, Where d represents the number of gradients included in the gradient. Based on the measurement of gradient contribution in this application study, D=3 is selected. When d=1, Containing only the coordinates of one of the first, second, or third gradients, when d=2, This includes the coordinates of the first and second gradients, or the first and third gradients, or the second and third gradients, when d=3. This includes the coordinates of three gradients: the first gradient, the second gradient, and the third gradient. The first gradient (G1), the second gradient (G2), and the third gradient (G3) are determined by spatial correlation calculations with the standard template gradients (G1_template, G2_template, G3_template). The correspondence is determined by the degree of correlation. The BrainSpace toolbox automatically performs the correlation calculations and outputs the corresponding G1, G2, and G3 values.

[0056] According to an embodiment of this application, the reference gradient space is the group average gradient space, based on the group to which the individual network's subject number ind belongs.

[0057] S102: Calculate the bulldozer distance between the gradient space of the first functional network and the gradient space of the second functional network to obtain the gradient spacing.

[0058] According to an embodiment of this application, the bulldozer distance quantifies the minimum cost required to transform from the gradient space of the first functional network to the gradient space of the second functional network.

[0059] Wasserstein Distance (optimal transmission distance), also known as Earth Mover's distance (EMD), is defined as the distance traveled by a vehicle. Distribution transformation distributed( Figure 5The minimum cost required to transform one pile of soil into the other (as shown in the diagram). If we imagine the distribution as two piles of soil with a certain amount of soil remaining, then EMD is the minimum amount of work (i.e., the minimum cost) required to transform one pile into the other. The amount of work is defined as the mass of a unit of soil multiplied by the distance it moves.

[0060] The distribution of two discrete mounds is denoted as and (like Figure 5 (As shown), the mound of earth Convert to There are many ways to bulldoze, and the goal of bulldozing is to find the one that minimizes the workload; this is an optimization problem. Based on Figure 5 The intuitive understanding shown is that it is a mound of earth. and A single bar in the column is denoted as and Each It refers to the amount of soil at the current position x. It refers to the final The amount of soil to be stored at a location, where the y-position is the same as the x-position, is used only to differentiate the distribution using different variables.

[0061] if : We need to remove the excess part at x ( The excavated earth was moved elsewhere; if We need to move some soil from other locations to location y, so that the earthwork stock at location y is [amount missing]. A relocation plan such as Figure 6 As shown, Figure 6 In It demonstrates dividing the mound of earth into multiple blocks. A relocation plan is presented, including the amount of movement and the distance to be moved.

[0062] pass Figure 5 , Figure 6 The intuitive understanding of bulldozing is further enhanced by defining the workload of earthmoving using mathematical symbols. Assuming a joint distribution... It is a joint distribution, and its marginal distribution is as mentioned above. and : .

[0063] It is the original distribution. It is the target distribution. It means that we need to move from point x. This amount of earthwork is transported to point y. The cost / price / cost of moving a unit of earthwork from point x to point y is defined as d(x,y). Common cost functions can generally be derived from the L-norm. For example: L1 norm ( L2 norm ( )wait.

[0064] Therefore, the Wasserstein distance can be defined as:

[0065]

[0066] inf represents the infimum, i.e., the minimum value. In other words, it means finding the total handling cost / price from all transportation options / transfer plans. Minimal transfer plan / transportation scheme The cost / price of this relocation plan is the bulldozer distance we need to calculate, expressed as... .

[0067] To explain this using a diagram: the shapes of the two mounds are determined ( and Determined), transportation costs / prices Determine the optimal transfer plan. The transportation cost / price is denoted as the Wasserstein distance between the two mounds. Therefore, the Wasserstein distance is also known as the "Earth Mover's Distance." The Earth Mover's Distance is the key factor in determining the cost of transporting the earth. The minimum value of , where It is definite, and this optimization needs to satisfy the following constraints: , and The constraint of the two marginal distributions can be understood as follows: the sum of the earthwork moved from point x (to all possible locations) must be equal to the original earthwork at point x. )quantity( The target earthwork after transportation is completed (). The existing earthwork volume at point y It must be equal to the sum of the earthwork from the original earthwork (at all possible locations). This means that the amount of earth moved is greater than or equal to zero. An integral is the limiting form of a summation, therefore it can be written in summation form, inner product form, and matrix form:

[0068]

[0069]

[0070]

[0071] like Figure 7 As shown, this is an example of a transfer quality. and transfer distance A schematic diagram of the matrix (where the transition distance is the transition cost), where It is a relocation plan A specific transfer plan in the diagram, where the value of an element represents the mass moved from position x to position y.

[0072] The constraints are expressed as follows:

[0073]

[0074]

[0075] Will and Two vectors are concatenated to form a longer vector b, and the two constraints are uniformly expressed as follows: .

[0076]

[0077] Therefore, the bulldozer distance is expressed as:

[0078]

[0079] According to embodiments of this application, in order to assess the distribution differences of each subject's functional network in gradient space, traditional methods typically estimate spatial separation by calculating the Euclidean distance between each subject and the network centroid of the group-level gradient template / gradient space. However, this method simplifies the entire brain network to a single representative point, ignoring spatial features such as shape, extent, and dispersion, resulting in information loss when quantifying the differences between individual subjects and the group-level gradient template. To address this issue, this application uses Earth Mover's Distance (EMD, also known as Wasserstein distance) instead of Euclidean distance to estimate spatial separation. This allows for the extraction of more information on the differences between individual subjects and the group-level gradient template through comparative analysis of structured point sets in gradient space. Compared to traditional Euclidean distance, applying EMD to the gradient analysis workflow integrates complete distribution information of all partitions within the network, providing more sensitive and geometrically meaningful cross-group comparison results.

[0080] According to embodiments of this application, based on individual gradient space Reference gradient space Construct the transition distance matrix / i.e., the transition cost matrix by calculating the Euclidean distance between intervals. The transfer cost can be calculated as the cost matrix for the entire gradient space (covering all gradients), or it can be calculated as the cost matrix for one or more specified gradients. The element in the i-th row and j-th column of the transfer cost matrix M represents the distance between the gradient coordinates of the i-th cortical partition in the individual gradient space and the gradient coordinates of the j-th cortical partition of the reference network (i.e., corresponding to d(x,y) in the mathematical expression), calculated as follows:

[0081]

[0082] Let be the element in the i-th row and j-th column of the cost matrix M. This represents the gradient coordinates of the j-th partition in the individual gradient space. This represents the gradient coordinates of the j-th partition of the reference gradient space. The gradient coordinates can cover all gradients, or they can specify only one or a few gradients. This represents the index of the specified gradient dimension; it is a set. A subset of the power set, Represents Euclidean distance. For example, sets The power set includes .

[0083] When calculating the cost matrix in only one dimension, The value is 1, at this time The difference between two coordinate values ​​is used when calculating the cost matrix in 2-dimensional and 3-dimensional dimensions. The distance between the coordinates is the Euclidean distance.

[0084] The bulldozer distance was quantified using two gradient spaces. and The EMD between spatial distributions is the optimal transformation cost (i.e., gradient spacing). This optimal transformation cost minimizes the total cost or total computation of transforming from one network distribution to another. To represent the offset between an individual and the reference gradient space, we define this transformation cost as the gradient spacing (also called network displacement) between the individual's gradient space and the reference network's gradient space. The network displacement is expressed as:

[0085]

[0086] in, Representing individual networks The index is The gradient space and reference network formed by the gradient coordinates The index is The distance between bulldozers in the gradient space formed by the gradient coordinates is simply expressed as... , Indicates from Convert to A collection of effective transfer plans, Represents any valid transfer plan in the set (a valid transfer plan is one that satisfies...). (A plan in which any element is greater than or equal to 0). yes The matrix inner product (Frobenius inner product) of M and M, which represents the total transformation cost, is calculated as follows:

[0087]

[0088] in, This represents the amount of mass that needs to be transferred from the i-th cortical partition of the personal network to the j-th cortical partition of the reference network. The element in the i-th row and j-th column of the transfer cost matrix M represents the distance between the gradient coordinates of the i-th cortical partition in the individual network and the gradient coordinates of the j-th cortical partition in the reference network. When there is only one gradient coordinate value, the distance is the absolute value of the difference. When there are multiple gradient coordinate values, the distance is the Euclidean distance.

[0089] For the spatial displacement of the functional network relative to the reference functional network, the gradient matrix of the functional network is used. Instead of the individual gradient space mentioned above , The default pattern network representing an individual contains the first... A gradient-dimensional network, and correspondingly, the reference network is represented as... .

[0090] For the distance between pairwise brain functional networks of an individual: the distance between brain functional networks is calculated by comparing the spatial distribution of each pair of brain functional networks at the individual level. For a given pair of functional networks... and ,in, or This represents a type of predefined functional network; for example, a seven-functional network. It can be any one of the following: DMN, SMN, VIS, SN, DAN, VAN, LIM.

[0091] We define a transition distance matrix (i.e., transition cost matrix) for a function transition. n is The number of cortical regions included, m is The number of cortical regions included.

[0092] Taking the seven functional networks of the brain as an example, since the number of cortical regions contained in each network is different (for example, the default mode network and the salience network contain different numbers of cortical regions), a partial matching strategy is adopted—only the total cortical mass of the network with fewer cortical regions is included. The network spacing based on the partial bulldozer distance representation is defined as:

[0093]

[0094] in, This represents the partial bulldozer distance between the i-th and j-th functional networks of an individual, and characterizes the network spacing between two functional networks of the individual. Let g represent the gradient coordinates of the i-th and j-th brain functional networks, respectively. Indicates the total number of transmission plans The feasible transfer plan space, taking the brain's seven functional networks as an example, is within the scope of the plan. .

[0095] This study uses the Python Optimal Transport Library (POT) to calculate the full and partial bulldozer distances (EMDs) using the ot.emd2 and ot.partial.partial_wasserstein functions, respectively. These calculations cover all combinations of gradient dimensions (1D, 2D, and 3D) (i.e., combinations within the power set) to ensure a comprehensive characterization of the network's geometric properties.

[0096] A second aspect of this application discloses a method for predicting post-traumatic stress disorder, the method comprising:

[0097] S201: Acquire functional magnetic resonance imaging of the subject;

[0098] S202: Based on the functional magnetic resonance imaging, perform functional connectivity analysis to obtain the brain functional connectivity matrix;

[0099] S203: Perform dimensionality reduction and gradient calculation on the brain functional connectivity matrix to obtain a low-dimensional gradient space, wherein the low-dimensional gradient space includes a first gradient, a second gradient, and a third gradient;

[0100] S204: Based on the low-dimensional gradient space, extract the gradient space of at least two functional networks on the set gradient to obtain the gradient space of the first functional network and the gradient space of the second functional network, wherein the set gradient is the third gradient.

[0101] S205: The gradient spacing is calculated based on the gradient space of the first functional network and the gradient space of the second functional network according to the brain functional gradient analysis method described above.

[0102] S206: Input the gradient spacing into the classifier to obtain the result of whether the subject suffers from post-traumatic stress disorder.

[0103] According to embodiments of this application, the following results were obtained in PTSD-related research: In this study, given that high-dimensional gradient combinations may introduce the curse of dimensionality and weaken interpretability, we focused on the first three gradients (which together explained approximately 55% of the variance) for the research dataset. In the HC, TEC, and PTSD groups, these three main axes stably characterize the macroscopic organization of the cortex, forming a common basis and principled framework for subsequent spatial analysis. Consistent with previous work, gradient 1 presents a main axis from the unimodal cortex (primary visual, somatosensory / motor) to the cross-modal association areas (medial prefrontal cortex, posterior cingulate cortex, inferior parietal cortex), corresponding to the classic hierarchy from extroverted sensory processing to introverted integration. Gradient 2 captures the sensory differentiation axis, primarily comparing the visual and auditory / somatosensory-motor cortices, suggesting modal separation within the unimodal cortex. Gradient 3 presents the control-default axis, distinguishing the preparietal system responsible for cognitive control from the default network, pointing to the functional division between goal-oriented operations and self-reference / introverted processing. Figure 8 ).

[0104] We first compared the explained variance of the three gradients with the range of the entire cortical network. The results showed significant group differences in the ranges of G1 and G3 (e.g., G1 had lower HC; G3 had lower PTSD: F and t tests, BF10, etc., are discussed in the main text), while the differences in range at the network level and within the network were not significant, suggesting that the global effect was not driven by a single network. At a finer-grained partitioning level, statistically significant partitions were mainly distributed in VIS (positive G1), SAN / VAN (negative G1), DMN (negative G2), and SAN / SMN (positive G2), while the inter-group differences in G3 were more extensive, involving SMN, DAN, SAN, FPN, and DMN; LIM was relatively stable across the three gradients. Overall, the common shifts related to trauma exposure were mainly reflected in G1 and G2, while PTSD-specific changes were more likely to be reflected along G3, pointing to complex perturbations in the balance between control-related systems and introverted systems.

[0105] Quantification of Gradient Space Functional Architecture: To refine the topological interpretation of the gradient space, we propose gradient spacing, which is the bulldozer distance or partial bulldozer distance between the gradient coordinates of functional networks. Specifically, it includes two types of indicators: "network displacement" (EMD) and "network spacing" (pEMD). Results show that in network displacement EMD, PTSD has a larger displacement relative to TEC on multiple networks: particularly seen in VIS, DAN, and SAN in G1; in SAN and DMN in G2; and in SMN, DAN, and FPN in G3. In 3D embedding, the group differences between VIS, DAN, and SAN are again observed. Regarding network spacing, the differences in G1 mainly involve the distance between VIS or SAN and other networks; G2 mainly involves the distance related to DMN and SAN; while G3 exhibits the most extensive reconstruction, covering almost all network pairs, and is mainly driven by the deviation of PTSD relative to HC / TEC. In 3D embedding, significant group differences are concentrated in SMN-DAN, SMN-SAN, SAN-FPN, SAN-DMN, and LIM-FPN. The summary of Euclidean distances by the zoning is highly consistent with the above findings: the reconstruction amplitude is larger along a single principal axis, while the global displacement is relatively convergent in the joint embedding space. The TEC and PTSD groups show significant differences from the HC group in Gradient 1 and Gradient 2, while the TEC and PTSD groups show almost no difference before that. The PTSD group shows significant differences from the HC and TEC groups in Gradient 3, while the HC group shows almost no difference from the TEC group.

[0106] Based on the differences between groups shown by the indicators calculated using gradient spacing, we propose to construct a PTSD prediction model based on the obtained indicators.

[0107] First, indicators showing significant differences between the PTSD group and the non-PTSD group include any one or more of the following: EMD, EMD(DMN), EMD(SMN), EMD(VIS), EMD(SN), EMD(DAN), EMD(VAN), EMD(LIM); pEMD(VIS_SMN), pEMD(VIS_DAN), pEMD(VIS_SAN), pEMD(VIS_LIM), pEMD(VIS_FPN), pEMD(VIS_DMN), pEMD(SMN_DAN), pEMD(SMN_SAN), pEMD(SMN_LIM), pEMD(SMN_FPN), pEMD(SMN_DMN), pEMD(DAN_SAN), pEMD(DAN_LIM), pEMD(DAN_FPN), pEMD(SAN_DMN), pEMD(SAN_FPN), pEMD(SAN_DMN). pEMD(LIM_FPN), pEMD(LIM_DMN), pEMD(FPN_DMN).

[0108] Based on the above-mentioned difference indicators, models can be built individually to distinguish PTSD, TEC, and HC, or models can be built in combination to obtain better predictive performance.

[0109] To verify the predictive power of the indicators discovered in this study for PTSD, we constructed a machine learning tri-class classification model using the calculated indicators. For the indicators obtained based on the third gradient, the performance of the corresponding indicators obtained by calculating the network spacing using centroid distance in existing techniques is shown in Table 1. When the centroid distance is replaced with the bulldozer distance used in this study to obtain pEMD, the predictive performance of pEMD for the three groups is shown in Table 2.

[0110] Table 1. Performance of Random Forests Based on Third Gradient of Centroid Distance Calculated as Network Spacing (CD)

[0111]

[0112] Table 2 Performance of Random Forest with Third Gradient Based on Network Spacing (pEMD) Calculated Using Partial Bulldozer Distance

[0113]

[0114] As can be seen, when using the three-class classification model, the correct hit rate for PTSD is improved to 0.75, and the F1-Score is improved from 0.72 to 0.81.

[0115] In other words, based on the results of this study, the above indicators can be used as an auxiliary diagnostic tool for PTSD: for new visitors, their fMRI data can be collected, their individual gradient space can be analyzed, and their displacement index and network spacing index can be calculated based on their individual gradient space and reference network. These can be input into a machine learning model to determine whether the visitor's imaging features match those of PTSD, thereby assisting psychiatrists or psychologists in making decisions and determining whether the visitor has PTSD.

[0116] Given the limited number of samples collected in this study, as the dataset is expanded, new data will be continuously incorporated to calibrate the reference network. Based on the calibrated reference network using large datasets, the aforementioned indicators will be calculated and the machine learning model trained, resulting in a more accurate PTSD-assisted diagnostic model. Among these, the most predictive inter-network shifts include: pEMD(VIS, SAN), pEMD(VIS, SMN), pEMD(VIS, DAN), pEMD(VIS, DMN), and pEMD(VIS, LIM). pEMD(VIS, SAN) represents the network distance between VIS and SAN, pEMD(VIS, SMN) represents the network distance between VIS and SMN, pEMD(VIS, DAN) represents the network distance between VIS and DAN, pEMD(VIS, DMN) represents the network distance between VIS and DMN, and pEMD(VIS, LIM) represents the network distance between VIS and LIM.

[0117] Because differentiating between TEC and PTSD groups in clinical practice is difficult, the predictive performance of CD as a predictive indicator for distinguishing between TEC and PTSD is poor. This application compares the predictive performance of using CD based on centroid distance and pEMD based on bulldozer distance to distinguish between TEC and PTSD. The predictive performance is shown in Table 3. Figure 9 As shown.

[0118] Table 3 shows the performance of classifiers built based on pEMD and CD in identifying TEC and PTSD.

[0119]

[0120] It is evident that the gradient spacing (network spacing) based on the improved bulldozer distance calculation can improve the discrimination level of PTSD and TEC.

[0121] Among them, the Top in Table 3 kThe indicators are the top k most important gradient distances between each pair of functional networks calculated using a seven-functional network. It can be seen that both pEMD and CD are distances between SMN and DAN. Meanwhile, the distances between FPN_DMN and SMN_FPN contribute significantly to both pEMD and CD, but their importance differs. This demonstrates that the bulldozer distance and the distance between the centroids capture different information, and the indicator based on the bulldozer distance captures more spatial features such as shape, range, and dispersion, thus more effectively distinguishing between PTSD and TEC. Table 3 shows the prediction indicators used in the model, and their importance is shown in Table 4.

[0122] Table 4. Corresponding Indicators of Classifiers

[0123]

[0124] According to an embodiment of this application, in S206, inputting the gradient spacing into the classifier to obtain the result of whether the subject suffers from post-traumatic stress disorder includes: constructing a three-classification model based on network spacing to distinguish PTSD, TEC, and HC; constructing a two-classification model based on network spacing to distinguish PTSD and non-PTSD (including TEC and HC); and constructing a two-classification model based on network spacing to distinguish PTSD and TEC (which has the highest clinical value and solves the problem of difficulty in identification in the prior art).

[0125] By examining the spatial gradient topological properties, we depicted the impact of trauma on large-scale functional network architecture. Spatially, TEC and PTSD showed consistent shifts along G1 and G2, primarily affecting the sensory system and DMN; while the changes in PTSD along G3 were more unique, suggesting a disruption of the separation between the control and default systems. At the "common" level, trauma-induced common changes spanned multiple dimensions in the gradient space: a rightward shift of the VIS along G1, a leftward shift of the DMN along G2, and a leftward shift of the FPN along G3. Previous literature has also widely reported similar network changes in PTSD and trauma populations, supporting the "shared changes" view after trauma. Notably, we did not observe significant changes in range and dispersion within the network—neither in regionalization nor at the network hierarchy—the gradient indices in the clinical group mainly reflected an overall "translation" of the network along the principal axis, rather than "extreme" changes in individual regionalizations within the network. Furthermore, the hierarchical structure of functional networks may lay the foundation for temporal organization: when a network deviates from its "typical" gradient position, its functional axes may overlap with those of other networks, thus "replacing" the latter's original interactions to some extent—a hypothesis we call the "spatial substitution hypothesis," which awaits further research. At the level of "difference," TEC and PTSD show extensive differentiation along G3, with most networks exhibiting displacements in opposite directions. G3 is often understood as the "task-resting / introverted-extroverted" cognitive dimension, related to processes such as attentional redirection, control signals, and executive functions. Therefore, dysfunction of higher-order cognitive systems may be the root cause of PTSD symptoms and constitute a major distinction from TEC. This perspective also helps explain why some indicators with significant group differences (such as network displacement and network spacing) are difficult to predict the severity of PTSD symptoms: the emergence of PTSD may be a "pathological crystallization" process from quantitative to qualitative change. Finally, our reflection on multidimensional gradient space is as follows: Evaluating the network's distribution across the entire space at the system level is undoubtedly reasonable; however, since each principal axis carries different functional dimensions, simply summarizing orthogonal dimensions may dilute the clear functional specificity changes on a single axis. Nevertheless, the overall geometry of multidimensional space may also contain deeper organizational patterns across axes, which warrants further exploration.

[0126] According to embodiments of this application, PTSD manifests as unique perturbations along gradient 3 and dynamic stabilization of maladaptive coactivation states, pointing to involvement of higher-order cognitive systems and potentially reflecting a pathological shift "from compensation to crystallization." The study proposes a novel gradient spatial distance quantification scheme, geometrically characterizing network displacement and separation (gradient spacing). This research provides a unified framework for understanding how trauma reshapes the brain's functional architecture and offers a basis for elucidating the bifurcation trajectories leading to clinical functional impairment. Embodiments of this application provide a computer device, which may include: one or more processors and one or more memories; wherein the memories store computer-readable code that, when executed by the one or more processors, can perform the methods described above.

[0127] The processor in this embodiment can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, operations, and logic block diagrams disclosed in this embodiment. The general-purpose processor can be a microprocessor or any conventional processor, and can be based on an x86 or ARM architecture.

[0128] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0129] For example, the method or apparatus according to embodiments of this disclosure can also be used by means of Figure 3 The architecture of the computing device 3000 shown is used for implementation. For example... Figure 3As shown, the computing device 3000 may include a bus 3010, one or more CPUs 3020, a read-only memory (ROM) 3030, a random access memory (RAM) 3040, a communication port 3050 connected to a network, an input / output component 3060, a hard disk 3070, etc. The storage devices in the computing device 3000, such as the ROM 3030 or the hard disk 3070, may store various data or files used for processing and / or communication of the methods provided in this disclosure, as well as program instructions executed by the CPU. The computing device 3000 may also include a user interface 3080. Of course, Figure 3 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 3 One or more components in the computing device shown.

[0130] This application also provides a computer-readable storage medium storing computer-readable instructions. When the computer-readable instructions are executed by a processor, the methods described with reference to the accompanying drawings according to embodiments of this disclosure can be performed. The computer-readable storage medium in the embodiments of this disclosure may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct memory bus random access memory (DR RAM). It should be noted that the memory used in the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0131] This disclosure also provides a computer program product or computer program that, when executed by a processor, implements the steps of the above-described method, such as... Figure 2As shown, the computer program product or computer program includes: a first acquisition module 301: used to acquire the gradient space of a first functional network and the gradient space of a second functional network, wherein the first functional network and the second functional network are the same functional networks in different brains or different functional networks in the same brain, and the functional network is a set including at least two functionally related brain regions; and a first gradient spacing calculation module 302: used to calculate the bulldozer distance between the gradient space of the first functional network and the gradient space of the second functional network to obtain the gradient spacing.

[0132] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0133] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0134] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0135] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0136] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0137] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0138] The exemplary embodiments of this disclosure described in detail above are merely illustrative and not restrictive. Those skilled in the art will understand that various modifications and combinations can be made to these embodiments or their features without departing from the principles and spirit of this disclosure, and such modifications should fall within the scope of this disclosure.

Claims

1. A method for analyzing brain functional gradients, characterized in that, The method includes: Obtain the gradient space of the first functional network and the gradient space of the second functional network, wherein the first functional network and the second functional network are the same functional networks in different brains or different functional networks in the same brain, and the functional network is a set including at least two functionally related brain regions. The gradient spacing is obtained by calculating the bulldozer distance between the gradient space of the first functional network and the gradient space of the second functional network. When the first functional network and the second functional network contain the same number of brain regions, the gradient spacing is obtained by calculating the bulldozer distance between the gradient spaces of the first functional network and the second functional network; the bulldozer distance is expressed as: ； in, Indicates the distance to the bulldozer. This represents the gradient space of the first functional network. This represents the gradient space of the second functional network. and Since they contain the same number of brain regions, min() represents finding the minimum value. Indicates from Convert to A collection of effective transfer plans, Represents any one of the valid transfer plans in the set of valid transfer plans. yes The inner product of the matrix M and M, where M is and The transition cost matrix between; When the first functional network and the second functional network contain different numbers of brain regions, the gradient spacing is obtained by calculating a partial bulldozer distance between the gradient spaces of the first functional network and the second functional network. This partial bulldozer distance is expressed as: ； in, This represents a partial bulldozer distance between the gradient space of the first functional network and the gradient space of the second functional network. This represents the gradient space of the first functional network. This represents the gradient space of the second functional network. and The number of brain regions included differs. M represents a feasible portion of the overall transfer plan. and The transition cost matrix between them yes The inner product of the matrix M and M, min() means finding the minimum value; The transfer cost matrix represents the transfer cost required to transform the gradient space of the first functional network into the gradient space of the second functional network. The element in the i-th row and j-th column of the transfer cost matrix represents the transfer cost of transforming the gradient coordinates of the i-th cortical partition in the gradient space of the first functional network into the gradient coordinates of the j-th cortical partition in the gradient space of the second functional network. The transfer cost represents the Euclidean distance between the gradient coordinates of the i-th cortical partition in the gradient space of the first functional network and the gradient coordinates of the j-th cortical partition in the gradient space of the second functional network. The element in the i-th row and j-th column of the transfer plan represents the transfer quality of converting the gradient coordinates of the i-th cortical partition in the gradient space of the first functional network to the gradient coordinates of the j-th cortical partition in the gradient space of the second functional network.

2. The brain functional gradient analysis method according to claim 1, characterized in that, The method for obtaining the gradient space of the first functional network includes: Acquire functional magnetic resonance imaging of the first subject; A first brain functional connectivity matrix was obtained by performing functional connectivity analysis based on the aforementioned functional magnetic resonance imaging. The first low-dimensional gradient space is obtained by performing dimensionality reduction and gradient calculation on the first brain functional connectivity matrix. The gradient space of the first functional network is obtained by extracting the coordinates on the set gradient of the first functional network in the first low-dimensional gradient space.

3. The brain functional gradient analysis method according to claim 2, characterized in that, The defined gradient is a set An element in the power set, wherein the defined gradient represents the gradient index in the low-dimensional gradient space, and D is the number of gradients contained in the low-dimensional gradient space.

4. A method for predicting post-traumatic stress disorder, characterized in that, The method includes: Acquire functional magnetic resonance imaging (fMRI) images of the subject; The brain functional connectivity matrix is ​​obtained by performing functional connectivity analysis based on the aforementioned functional magnetic resonance imaging. The brain functional connectivity matrix is ​​reduced in dimensionality and gradient is calculated to obtain a low-dimensional gradient space, which includes a first gradient, a second gradient, and a third gradient. Based on the low-dimensional gradient space, at least two functional networks are extracted to obtain the gradient space of the first functional network and the gradient space of the second functional network, wherein the set gradient is the third gradient. The gradient spacing is calculated based on the gradient space of the first functional network and the gradient space of the second functional network according to any one of claims 1-3. The gradient spacing is input into the classifier to obtain the result of whether the subject suffers from post-traumatic stress disorder.

5. The method for predicting post-traumatic stress disorder according to claim 4, characterized in that, The gradient spacing includes one or more of the following: gradient spacing between the default mode network and the visual network; gradient spacing between the default mode network and the salience network; gradient spacing between the default mode network and the dorsal attention network; gradient spacing between the default mode network and the ventral attention network; gradient spacing between the default mode network and the sensorimotor network; gradient spacing between the default mode network and the edge network; gradient spacing between the visual network and the salience network; gradient spacing between the visual network and the dorsal attention network; gradient spacing between the visual network and the ventral attention network; gradient spacing between the visual network and the sensorimotor network; gradient spacing between the visual network and the edge network; gradient spacing between the salience network and the dorsal attention network; gradient spacing between the salience network and the ventral attention network; gradient spacing between the salience network and the sensorimotor network; gradient spacing between the salience network and the edge network; gradient spacing between the dorsal attention network and the ventral attention network; gradient spacing between the dorsal attention network and the sensorimotor network; gradient spacing between the dorsal attention network and the edge network; gradient spacing between the ventral attention network and the sensorimotor network; gradient spacing between the ventral attention network and the edge network; gradient spacing between the sensorimotor network and the edge network.

6. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1-5.

7. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-5.

Images