A brain disease classification method and system

Abstract

The application discloses a brain disease classification method and system, which realizes accurate diagnosis through multi-modal data fusion and dynamic modeling. The method comprises the following steps: collecting multi-modal brain image information and cognitive behavior information of a user; performing dynamic functional connection analysis on resting-state functional magnetic resonance timing signals to obtain a time-varying brain network characteristic matrix, and simultaneously performing white matter fiber bundle topological reconstruction on a structural connection matrix; constructing a four-dimensional correlation tensor through a neural dynamics model by using the time-varying network characteristics, the structural connection weight and the anatomical characteristics; performing multi-task learning on the four-dimensional correlation tensor based on a time-varying graph neural network model to output a quantitative diagnosis result; and finally generating a clinical classification report integrating individualized brain network remodeling target points, disease progression risk stratification and treatment response prediction. Through dynamic fusion of structural and functional characteristics, the application realizes comprehensive characterization of the pathological mechanism of brain diseases, and provides a decision support with accuracy and interpretability for clinical diagnosis.

Description

Technical Field

[0001] This invention relates to the field of medical image analysis technology, specifically to a method and system for classifying brain diseases. Background Technology

[0002] Clinical diagnosis of brain diseases primarily relies on a combination of medical imaging examinations and neuropsychological assessments. Currently used diagnostic methods are typically based on single types of brain scan data, such as analyzing morphological changes in specific brain regions using structural magnetic resonance imaging (MRI), or calculating the correlations between different brain regions using functional imaging data. While these methods can identify some abnormal features, they struggle to fully reflect the complex pathological mechanisms of brain diseases.

[0003] In functional analysis, conventional methods often treat the functional connections between brain regions as fixed and unchanging, ignoring the dynamic changes in neural activity. Furthermore, techniques for analyzing brain structural connectivity typically focus only on the physical connectivity characteristics of white matter fibers, failing to adequately consider their interaction with neural functional activity. Most existing computer-aided diagnostic systems employ traditional classification algorithms, which have limited ability to integrate multi-source data and typically output only simple classification labels, lacking explanations of the underlying biological mechanisms of diseases.

[0004] With the development of medical technology, clinical practice has placed higher demands on the diagnosis of brain diseases, requiring novel analytical methods that can integrate multi-dimensional information, reflect the dynamic characteristics of diseases, and provide interpretable results. This need is particularly prominent in the diagnosis of neurological diseases with complex pathological mechanisms, such as neurodegenerative diseases and mental disorders. Summary of the Invention

[0005] In view of the above problems, the present invention provides a brain disease classification method and system, which constructs a time-varying neural network model by dynamically fusing multimodal brain imaging and cognitive behavioral data, thereby achieving accurate classification and mechanism explanation and solving the problem of difficulty in comprehensively characterizing the dynamic pathological features of brain diseases.

[0006] To achieve the above objectives, in a first aspect, this application provides a method for classifying brain diseases, comprising:

[0007] Multimodal brain imaging and cognitive behavioral information of users are collected. The multimodal brain imaging information includes resting-state functional magnetic resonance imaging time-series signals, diffusion tensor imaging structural connectivity matrix and T1-weighted anatomical features. Cognitive behavioral information is obtained through standardized neuropsychological scales.

[0008] Dynamic functional connectivity analysis was performed on the resting-state functional magnetic resonance imaging time-series signal to obtain the time-varying brain network feature matrix. The dynamic functional connectivity analysis was configured to perform phase synchronization calculation based on a sliding window, and white matter fiber bundle topology reconstruction was performed on the diffusion tensor imaging structural connectivity matrix to obtain the structural connectivity weight matrix.

[0009] The time-varying brain network feature matrix, structural connectivity weight matrix, and T1-weighted anatomical features are used to construct a four-dimensional correlation tensor through a neurodynamic model. The dimensions of the four-dimensional correlation tensor include spatial dimension, temporal dimension, connectivity strength dimension, and gray matter morphology dimension.

[0010] Multi-task joint learning of the four-dimensional correlation tensor is performed based on the time-varying graph neural network model, and the output is a quantitative diagnostic result containing disease classification probability, abnormal markers of neural circuits and cognitive decline trajectory. The time-varying graph neural network model is configured as a coupled oscillator embedding module with biophysical constraints.

[0011] Based on the cross-modal alignment of quantitative diagnostic results with cognitive behavioral information, an auxiliary clinical classification report is generated, which includes individualized brain network remodeling targets, disease progression risk stratification, and treatment response prediction.

[0012] In some embodiments, dynamic functional connectivity analysis is performed on resting-state functional magnetic resonance imaging (fMRI) time-series signals to obtain a time-varying brain network feature matrix. The dynamic functional connectivity analysis is configured to perform phase synchronization calculation based on a sliding window, including:

[0013] Sliding window analysis was performed on the resting-state functional magnetic resonance imaging time-series signal to obtain multiple sliding windows. The sliding window analysis was configured to be performed using a sliding window algorithm based on adaptive window length. The window length of the sliding window was dynamically adjusted according to the signal-to-noise ratio of multiple brain region signals.

[0014] The phase synchronization intensity of the brain region corresponding to each sliding window is calculated and expressed by formula (1), which is as follows:

[0015] ;

[0016] In formula (1), for Time brain region With brain regions The strength of phase synchronization between them The total number of time points within a single sliding window. brain region for The instantaneous phase at a given time point brain region for The instantaneous phase at a given time point To calculate brain regions With brain regions The complex exponential function of the phase difference vector between them. For counting letters, ;

[0017] Repeat the above steps until the phase synchronization intensity of all brain regions has been calculated, and construct a two-dimensional dynamic functional connectivity matrix based on the phase synchronization intensity, expressed by formula (2), which is as follows:

[0018] ;

[0019] In formula (2), For dynamic functional connection matrix;

[0020] The dynamic functional connectivity matrix is ​​reduced in dimension by nonnegative matrix factorization, and the time-varying brain network feature matrix is ​​extracted, which is expressed by formula (3) as follows:

[0021] ;

[0022] In formula (3), These are the static network components after nonnegative matrix decomposition. This is the feature matrix of a time-varying brain network.

[0023] In some embodiments, white matter fiber bundle topological reconstruction is performed on the diffusion tensor imaging structure connectivity matrix to obtain a structure connectivity weight matrix, including:

[0024] The probabilistic fiber tracing algorithm was used to reconstruct the whole-brain white matter fiber pathways from the diffusion tensor imaging structural connectivity matrix, and the initial structural connectivity strength of the brain region was calculated based on the spatial density distribution of fiber bundles.

[0025] The initial structure connectivity matrix is ​​standardized using graph theory measures to obtain the structure connectivity matrix, which is expressed by formula (4), as follows:

[0026] ;

[0027] In formula (4), brain region With brain regions The initial structural connection strength between them The average anisotropy fraction of the fiber bundle. To determine the number of connecting fibers, The average fiber length, brain region With brain regions The distance between their centroids The width of the Gaussian kernel;

[0028] The structural connection matrix is ​​represented by formula (5), which is as follows:

[0029] ;

[0030] In formula (5), For structural connection matrix;

[0031] By performing a diffusion geometry mapping on the structural connectivity matrix, a low-dimensional manifold space is obtained;

[0032] Spectral clustering algorithm is used to identify features in low-dimensional manifold space to obtain modular features, which include intra-module connectivity enhancement coefficients and inter-module connectivity attenuation coefficients.

[0033] The structural connection matrix is ​​weighted based on modular characteristics, and the structural connection weight matrix is ​​obtained by adjusting the intra-module connection enhancement coefficient and inter-module connection attenuation coefficient using two parameters.

[0034] In some embodiments, a probabilistic fiber tracing algorithm is used to reconstruct whole-brain white matter fiber pathways from the diffusion tensor imaging structural connectivity matrix, and the initial structural connectivity strength between brain regions is calculated based on the spatial density distribution of fiber bundles, including:

[0035] In the anisotropic field defined by the structure connection matrix of the diffusion tensor imaging, the propagation path of white matter fibers is simulated by Monte Carlo random walk. The propagation path of white matter fibers in each simulation is obtained by probability sampling based on the direction of the principal eigenvector of the local diffusion tensor.

[0036] Spatial clustering is performed on the fiber bundle paths generated by repeated sampling, and the connected paths passing through the preset white matter skeleton region are retained and recorded as the initial fiber bundle paths;

[0037] The cross-distribution density of the initial fiber tract paths between various brain regions is statistically analyzed, and geometric correction is performed based on the length of the initial fiber tract paths to generate an initial structural connectivity matrix with directional weights.

[0038] In some embodiments, spectral clustering algorithms are used to identify features in a low-dimensional manifold space to obtain modular features, including:

[0039] Calculate multiple eigenvectors based on the Laplacian matrix in a low-dimensional manifold space, and construct an eigenvector space;

[0040] The k-means clustering algorithm is used to partition the feature vector space, resulting in modular partitions of the brain network;

[0041] Calculate the membership degree of each brain region in the modular partition of the brain network to generate a modular feature matrix;

[0042] The structural connection matrix is ​​weighted based on modular characteristics, and adjusted using a dual-parameter method that combines intra-module connection enhancement coefficients and inter-module connection attenuation coefficients to obtain the structural connection weight matrix, which includes:

[0043] An enhancement factor is applied to the connectivity of brain regions belonging to the same module to increase the strength of intra-module connectivity; this factor is denoted as the intra-module connectivity enhancement factor.

[0044] An attenuation coefficient is applied to cross-module brain region connections to suppress the strength of inter-module connections; this coefficient is denoted as the inter-module connection attenuation coefficient.

[0045] Repeat the above steps until all brain regions have been traversed;

[0046] The intra-module connection enhancement coefficients and inter-module connection attenuation coefficients are normalized to generate a structural connection weight matrix with modular topological constraints.

[0047] In some embodiments, a four-dimensional correlation tensor is constructed using a neurodynamic model by combining the time-varying brain network feature matrix, the structural connectivity weight matrix, and the T1-weighted anatomical features, including:

[0048] A multimodal feature fusion framework is constructed to map the dynamic phase synchronization mode of the time-varying brain network feature matrix to the energy transfer equation of the neural mass unit;

[0049] A white matter conduction delay constraint is imposed on the structural connectivity weight matrix, and it is coupled to the spatial propagation term of the neural mass unit through a diffusion-reaction equation.

[0050] Cortical thickness and gray matter density distribution were extracted from T1-weighted anatomical features and used as local excitability regulation parameters for neural quality units.

[0051] By integrating the energy transfer equation, spatial propagation term, and excitability regulation parameters through tensor product operations, a four-dimensional correlation tensor with spatiotemporal, connectivity, and morphological characteristics is generated. The spatial dimension is configured as the spatial topological arrangement of neural mass units, the temporal dimension is configured as the oscillation period of dynamic phase synchronization, the connectivity strength dimension is configured as the synaptic weights of white matter conduction, and the gray matter morphological dimension is configured as the changes in cortical thickness gradient.

[0052] In some embodiments, the dynamic phase synchronization mode of the time-varying brain network feature matrix is ​​mapped to the energy transfer equation of the neural mass unit, which is expressed by formula (6), as follows:

[0053] ;

[0054] In formula (6), The total number of brain regions, Indicates the first The energy state of each neural mass unit This represents the normalized functional connectivity weights of the time-varying brain network feature matrix. Indicates the first The oscillation phase of each neural mass unit Indicates the first The oscillation phase of each neural mass unit The energy decay coefficient, The intensity of the external input stimulus;

[0055] It is coupled to the spatial propagation term of the neural mass unit through the diffusion-reaction equation, and is expressed by Equation (7), which is as follows:

[0056] ;

[0057] In formula (7), The neuron membrane time constant, For the first The activity potential of a neural mass unit This represents the structural connection weights in the structural connection matrix. For delay time variables, This indicates the white matter conduction delay time in brain regions. It is the Sigmoid activation function. This is a Gaussian noise term;

[0058] By integrating the energy transfer equation, spatial propagation term, and excitability regulation parameter through tensor product operation, a four-dimensional correlation tensor with spatiotemporal-connectivity-morphology four-dimensional characteristics is generated, which is expressed by formula (8), as follows:

[0059] ;

[0060] In formula (8), Represents a four-dimensional correlation tensor. Represents a neural mass unit. Representation of neural mass unit The target neural mass unit of structural connectivity. Represents the time series of energy states. This represents the discretization level of the white matter conduction delay time. Indicates the discretization level The representative value, This represents the tensor product operation. Indicates the first The normalized value of the product of gray matter density and cortical thickness in each brain region.

[0061] In some embodiments, a multi-task joint learning of a four-dimensional correlation tensor is performed based on a time-varying graphical neural network model to output a quantitative diagnostic result containing disease classification probability, neural circuit abnormality markers, and cognitive decline trajectory, including:

[0062] A biophysically constrained coupled oscillator module is constructed in a time-varying graphical neural network model, including:

[0063] The features contained in the time dimension of the four-dimensional correlation tensor are mapped to the oscillator phase dynamics equation. The natural frequency of each oscillator unit is determined by the resting-state EEG feature spectrum, and the coupling strength is dynamically adjusted by the connection strength dimension in the four-dimensional correlation tensor.

[0064] The instantaneous phase synchronization matrix of the oscillator cluster is calculated using a phase-locked detection algorithm and used as the dynamic connection feature input of the graph neural network.

[0065] Spatiotemporal graph convolution based on anatomical constraints processes dynamic connectivity features. Spatiotemporal graph convolution includes spatial dimension convolution processing and temporal dimension convolution processing. Spatial dimension convolution processing includes spatial graph convolution, and the adjacency matrix of spatial graph convolution is limited to the white matter connectivity path determined by DTI fiber tracing. Temporal dimension convolution processing includes deformable convolution kernels to capture multi-scale features from event-related potentials to functional state fluctuations.

[0066] In the convolution operation of each spatiotemporal graph convolution, the node features and connection weights are updated synchronously until the processing is complete;

[0067] Quantitative diagnostic results are generated through a multi-task attention mechanism, including:

[0068] Multitasking attention mechanisms include disease classification branches, neural circuit abnormality monitoring branches, and cognitive prediction branches;

[0069] The disease classification branch calculates the contribution weight of each brain region to the classification decision through dynamic graph pooling, and obtains the first quantitative result containing the first attention weight;

[0070] The neural circuit anomaly detection branch identifies abnormal pathways where the functional connectivity strength and structural connectivity density deviate significantly through side attention scores, and obtains a second quantification result containing a second attention weight.

[0071] The cognitive prediction branch uses a time-attention algorithm to calculate the fluctuation cycle and obtains a third quantization result containing a third attention weight;

[0072] A quantitative diagnostic result is generated based on the first, second, and third quantitative results.

[0073] In some embodiments, multimodal brain imaging information and cognitive behavioral information of the user are collected. The multimodal brain imaging information includes resting-state functional magnetic resonance imaging time-series signals, diffusion tensor imaging structural connectivity matrices, and T1-weighted anatomical features, including:

[0074] The resting-state functional magnetic resonance time-series signal was configured to be acquired using a preset functional magnetic resonance imaging sequence;

[0075] The diffusion tensor imaging structure connection matrix is ​​configured to acquire data via a preset diffusion-weighted imaging protocol;

[0076] T1-weighted anatomical features were configured to be acquired using high-resolution three-dimensional structural imaging sequences, and the brain was divided into a preset number of automatically anatomically labeled brain regions using an atlas-based segmentation method.

[0077] In a second aspect, the present invention also provides a brain disease classification system applicable to the brain disease classification method described in the first aspect. The system includes a data acquisition module, a data processing module, a logic operation module, and a result output module. The data acquisition module is used to collect multimodal brain imaging information and cognitive behavioral information from users. The multimodal brain imaging information includes resting-state functional magnetic resonance imaging (fMRI) time-series signals, diffusion tensor imaging (DTI) structural connectivity matrices, and T1-weighted anatomical features. The cognitive behavioral information is obtained through standardized neuropsychological scales. The data processing module is used to perform dynamic functional connectivity analysis on the resting-state fMRI time-series signals to obtain a time-varying brain network feature matrix. The dynamic functional connectivity analysis is configured to perform phase synchronization calculation based on a sliding window, and to perform white matter fiber tract topological reconstruction on the DTI structural connectivity matrix to obtain… The structural connectivity weight matrix and the logical operation module are used to construct a four-dimensional correlation tensor by combining the time-varying brain network feature matrix, the structural connectivity weight matrix, and T1-weighted anatomical features through a neurodynamic model. The dimensions of the four-dimensional correlation tensor include spatial dimension, temporal dimension, connectivity strength dimension, and gray matter morphology dimension. Based on the time-varying graphical neural network model, the four-dimensional correlation tensor is subjected to multi-task joint learning to output quantitative diagnostic results containing disease classification probability, neural circuit abnormality markers, and cognitive decline trajectory. The time-varying graphical neural network model is configured as a coupled oscillator embedding module with biophysical constraints. The result output module is used to generate an auxiliary clinical classification report containing individualized brain network remodeling targets, disease progression risk stratification, and treatment response prediction based on the cross-modal alignment degree between the quantitative diagnostic results and cognitive behavioral information.

[0078] Unlike existing technologies, the above-mentioned technical solution provides a method and system for classifying brain diseases. It achieves accurate diagnosis through multimodal data fusion and dynamic modeling. The method includes: collecting multimodal brain imaging information and cognitive behavioral information from users, wherein the multimodal brain imaging information includes resting-state functional magnetic resonance imaging (fMRI) time-series signals, diffusion tensor imaging (DTI) structural connectivity matrices, and T1-weighted anatomical features; performing dynamic functional connectivity analysis on the resting-state fMRI time-series signals to obtain a time-varying brain network feature matrix, and simultaneously reconstructing the white matter fiber tract topology of the structural connectivity matrix; constructing a four-dimensional correlation tensor using a neurodynamic model by combining the time-varying network features, structural connectivity weights, and anatomical features; performing multi-task learning on the four-dimensional correlation tensor based on a time-varying graphical neural network model to output quantitative diagnostic results including disease classification probabilities, neural circuit abnormality markers, and cognitive decline trajectories; and finally generating a clinical classification report integrating personalized brain network remodeling targets, disease progression risk stratification, and treatment response prediction. This invention, through the dynamic fusion of structural and functional features, achieves a comprehensive characterization of the pathological mechanisms of brain diseases, providing decision support for clinical diagnosis that is both accurate and interpretable.

[0079] The above description of the invention is merely an overview of the technical solution of this application. In order to enable those skilled in the art to better understand the technical solution of this application and to implement it based on the description and drawings, and to make the above-mentioned objectives and other objectives, features and advantages of this application easier to understand, the following description is provided in conjunction with the specific embodiments and drawings of this application. Attached Figure Description

[0080] The accompanying drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of specific embodiments of the present invention and other related contents, and should not be considered as limitations on this application.

[0081] In the accompanying drawings of the instruction manual:

[0082] Figure 1 This is a flowchart illustrating steps S101 to S105 of the classification method described in a specific implementation.

[0083] Figure 2 This is a flowchart illustrating steps S201 to S203 of the classification method described in a specific implementation.

[0084] Figure 3 This is a schematic diagram of the classification system described in a specific implementation.

[0085] The reference numerals used in the above figures are explained as follows:

[0086] 1. Classification system;

[0087] 11. Data acquisition module;

[0088] 12. Data processing module;

[0089] 13. Logical operation module;

[0090] 14. Results Output Module. Detailed Implementation

[0091] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0092] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0093] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0094] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0095] In this application, terms such as “first” and “second” are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy or order relationship between these entities or operations.

[0096] Without further limitations, the use of terms such as “comprising,” “including,” “having,” or other similar open-ended expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0097] As understood in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments in this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0098] In the description of the embodiments of this application, the space-related expressions used, such as "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or drawings. They are only for the purpose of describing the specific embodiments of this application or for the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0099] The processor described in the embodiments of this application can be implemented by hardware, firmware, software, or a combination thereof. It can be a circuit, one or more of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a central processing unit (CPU), a controller, a microcontroller, or a microprocessor. It also includes other physical, biological, or chemical structures that can implement the same or equivalent functions as the processors listed above, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute some or all of the steps in the computer program or method involved in the various embodiments of this application, or any combination of the steps mentioned therein.

[0100] The computer program involved in the embodiments can be stored in a computer device readable storage medium, which includes, but is not limited to, disks, magnetic tapes, magnetic cards, floppy disks, flash memory, optical disks, optical cards, read-only memory (ROM), random access memory (RAM), erasable programmable ROM (EPROM), and electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical, or chemical structures that can achieve the same or equivalent functions as the storage media listed above, such as DNA, RNA, proteins, and other units with information storage capabilities. In specific embodiments, the storage medium involved can be one of the above-mentioned media types, or a combination of the above-mentioned media types. In different embodiments, the computer program involved in the embodiments can be centrally stored in a single medium, or distributed and stored in multiple media. The memory containing the computer device readable storage medium can be non-volatile memory or random access memory. These computer device readable storage media can be built into the device, or can be connected to the device involved in the embodiments as an external device or part of an external device. In some embodiments, the memory having a computer device readable storage medium is deployed locally; in other embodiments, the memory may be deployed remotely from the processor, for example, as a network-attached memory accessed via RF circuitry or an external port and a communication network, wherein the communication network may be the Internet, one or more intranets, a local area network (LAN), a wide area network (WLAN), a storage area network (SAN), or a suitable combination thereof, as long as computer device access to the memory is enabled. Furthermore, the computer program involved in the embodiments may be stored in plaintext / ciphertext form, or it may be designed as training data, integrated and recombined through model training and implicitly stored in the parameter states of a deep neural network or other machine learning model.

[0101] Please see Figure 1 In a first aspect, this embodiment provides a method for classifying brain diseases, including:

[0102] S101. Collect multimodal brain imaging information and cognitive behavioral information of users. Multimodal brain imaging information includes resting-state functional magnetic resonance imaging time-series signals, diffusion tensor imaging structural connectivity matrix and T1-weighted anatomical features. Cognitive behavioral information is obtained through standardized neuropsychological scales.

[0103] S102. Dynamic functional connectivity analysis is performed on the resting-state functional magnetic resonance imaging time-series signal to obtain the time-varying brain network feature matrix. The dynamic functional connectivity analysis is configured to perform phase synchronization calculation based on a sliding window, and white matter fiber bundle topological reconstruction is performed on the diffusion tensor imaging structural connectivity matrix to obtain the structural connectivity weight matrix.

[0104] S103. The time-varying brain network feature matrix, structural connection weight matrix and T1 weighted anatomical features are combined to construct a four-dimensional correlation tensor through a neurodynamic model. The dimensions of the four-dimensional correlation tensor include spatial dimension, temporal dimension, connection strength dimension and gray matter morphology dimension.

[0105] S104. Based on the time-varying graph neural network model, multi-task joint learning is performed on the four-dimensional correlation tensor to output quantitative diagnostic results including disease classification probability, abnormal markers of neural circuits and cognitive function decline trajectory. The time-varying graph neural network model is configured as a coupled oscillator embedding module with biophysical constraints.

[0106] S105. Generate an auxiliary clinical classification report that includes individualized brain network remodeling targets, disease progression risk stratification, and treatment response prediction based on the cross-modal alignment of quantitative diagnostic results and cognitive behavioral information.

[0107] In step S101, the acquisition of multimodal brain imaging information is completed through medical imaging equipment. Among them, the resting-state functional magnetic resonance imaging time-series signal reflects the dynamic functional activity of brain regions and can be acquired using a high temporal resolution scanning protocol to ensure that transient features of neural activity can be captured; the diffusion tensor imaging structural connectivity matrix characterizes the physical connectivity features of white matter fiber bundles and can be reconstructed through a multi-b-value sampling scheme to improve the resolution of crossing fibers; T1-weighted anatomical features are used to quantify morphological indicators such as gray matter volume in brain regions, and isotropic voxel scanning is preferred to provide high spatial accuracy data for subsequent morphological analysis; cognitive behavioral information is acquired through standardized neuropsychological scale assessment, including cognitive domain scores such as memory and executive function. The scale selection is determined based on the core cognitive deficit domain of the target disease to ensure the biological correlation between behavioral assessment and imaging features.

[0108] In step S102, dynamic functional connectivity analysis uses an adaptive sliding window technique to process resting-state functional magnetic resonance imaging time-series signals. By calculating the phase synchronicity of brain regions within the window, the time-varying characteristics of functional connectivity are captured, ultimately forming a time-varying brain network feature matrix that reflects the dynamic changes of the brain network. The window length is dynamically adjusted according to the autocorrelation characteristics of the signal, ensuring both the precision of the temporal resolution and maintaining sufficient statistical reliability. Preferably, the phase synchronicity calculation uses a weighted phase lag exponent to effectively suppress spurious connections caused by volume conduction effects.

[0109] Simultaneously, white matter fiber tract topological reconstruction is performed on the diffusion tensor imaging structural connectivity matrix. Topological indices such as node centrality are calculated using graph theory methods to generate a structural connectivity weight matrix. Preferably, the white matter fiber tract topological reconstruction process incorporates an adaptive threshold strategy based on streamline density. By analyzing the distribution characteristics of connectivity strength, the saliency threshold of structural connectivity is automatically determined, avoiding subjective bias caused by manual setting. This step extracts key features of the brain network from both functional and structural perspectives.

[0110] In step S103, the neurodynamic model performs multimodal fusion of the time-varying brain network feature matrix, structural connectivity weight matrix, and T1-weighted anatomical features to construct a four-dimensional correlation tensor. The spatial dimension corresponds to brain region location, the temporal dimension records dynamic changes in functional connectivity, the connectivity strength dimension represents structural connectivity weights, and the gray matter morphology dimension stores anatomical features. This tensor structure achieves spatiotemporal alignment of different modal features within a unified framework.

[0111] Preferably, the neurodynamic model adopts a coupled differential equation system, which mathematically models the dynamic change law of the time-varying brain network feature matrix and the constraint conditions of the structural connection weight matrix. The coupling strength parameter is determined by fitting actual neuroelectrophysiological data to ensure that the model has a real biophysical basis. The construction process of the four-dimensional correlation tensor adopts tensor decomposition technology, which retains the most discriminative feature components in each modality data through low-rank approximation, while realizing data dimensionality reduction and feature fusion.

[0112] In step S104, the time-varying graph neural network model simulates the biophysical characteristics of neural activity through a coupled oscillator embedding module, performing multi-task joint learning on the four-dimensional correlation tensor. The model simultaneously outputs three types of results: disease classification probability, neural circuit abnormality markers, and cognitive function decline trajectory. The disease classification probability reflects the confidence level of the diagnostic conclusion, neural circuit abnormality markers indicate dysfunction of specific brain networks, and the cognitive function decline trajectory predicts the cognitive change trend in disease progression.

[0113] Preferably, the inherent frequency parameters of the coupled oscillator embedded module are personalized according to the metabolic characteristics of each brain region; the attention weights are dynamically adjusted through the time-varying mode of the learning function connection; the multi-task joint learning adopts a hierarchical sharing mechanism, with the bottom layer network sharing the extraction of basic neural features, and the upper layer network being professionally adjusted for classification tasks, anomaly labeling tasks and trajectory prediction tasks respectively, thereby improving the performance of each task while ensuring model efficiency.

[0114] In step S105, preferably, a cross-modal Bayesian inference framework is established to achieve deep alignment between quantitative diagnostic results and cognitive behavioral information. In the generated auxiliary clinical classification report, the determination of individualized brain network remodeling targets comprehensively considers the spatial distribution intensity and temporal stability of abnormal markers; the threshold for disease progression risk stratification is set based on statistical regularities from large-sample longitudinal studies; and treatment response prediction is assessed by combining pharmacodynamic models and the remodeling potential of functional networks. This multi-dimensional report not only provides decision-making suggestions but also includes credibility indicators for various conclusions, assisting physicians in weighing the risks and benefits.

[0115] This embodiment captures the dynamic characteristics of brain networks through temporal signal analysis and overcomes the limitations of single data sources by utilizing multimodal fusion, ultimately achieving cross-scale correlation analysis from microscopic neural circuit abnormalities to macroscopic cognitive and behavioral changes. Data acquisition ensures information integrity, feature extraction preserves biological characteristics, and model construction achieves multi-dimensional integration, ultimately outputting clinical application results that balance classification accuracy and mechanism explanation.

[0116] In some embodiments, dynamic functional connectivity analysis is performed on resting-state functional magnetic resonance imaging (fMRI) time-series signals to obtain a time-varying brain network feature matrix. The dynamic functional connectivity analysis is configured to perform phase synchronization calculation based on a sliding window, including:

[0117] Sliding window analysis was performed on the resting-state functional magnetic resonance imaging time-series signal to obtain multiple sliding windows. The sliding window analysis was configured to be performed using a sliding window algorithm based on adaptive window length. The window length of the sliding window was dynamically adjusted according to the signal-to-noise ratio of multiple brain region signals.

[0118] The phase synchronization intensity of the brain region corresponding to each sliding window is calculated and expressed by formula (1), which is as follows:

[0119] ;

[0120] In formula (1), for Time brain region With brain regions The strength of phase synchronization between them The total number of time points within a single sliding window. brain region for The instantaneous phase at a given time point brain region for The instantaneous phase at a given time point To calculate brain regions With brain regions The complex exponential function of the phase difference vector between them. For counting letters, ;

[0121] Repeat the above steps until the phase synchronization intensity of all brain regions has been calculated, and construct a two-dimensional dynamic functional connectivity matrix based on the phase synchronization intensity, expressed by formula (2), which is as follows:

[0122] ;

[0123] In formula (2), For dynamic functional connection matrix;

[0124] The dynamic functional connectivity matrix is ​​reduced in dimension by nonnegative matrix factorization, and the time-varying brain network feature matrix is ​​extracted, which is expressed by formula (3) as follows:

[0125] ;

[0126] In formula (3), These are the static network components after nonnegative matrix decomposition. This is the feature matrix of a time-varying brain network.

[0127] In this embodiment, when resting-state functional magnetic resonance imaging (fMRI) time-series signals are analyzed and processed through a sliding window, an adaptive window length algorithm dynamically adjusts the window length based on the signal-to-noise ratio (SNR) of each brain region. The SNR is calculated by comparing the signal power of the target brain region with the background noise power. When high-frequency neural activity components are detected, the window is automatically shortened to improve temporal resolution, while in the low-frequency band where signal stability is relatively high, the window is appropriately lengthened to ensure statistical reliability. This adaptive mechanism overcomes the limitations of a fixed window length in capturing neural activity at different frequency bands.

[0128] Phase synchronization strength is calculated using the Phase Lock Value (PLV), which quantifies the degree of functional coupling between brain regions by calculating the mean amplitude of the complex exponential function of the instantaneous phase difference vector within a statistical sliding window. Instantaneous phase extraction is achieved using a Hilbert transform to ensure accurate reflection of the temporal characteristics of neural oscillations. For a single sliding window, all brain region pairs (e.g., brain region...) are considered. With brain regions After the phase synchronization strength is calculated, the matrix is ​​arranged in rows and columns to form a dynamic functional connection matrix. Its diagonal elements characterize the spontaneous activity stability of each brain region, while the off-diagonal elements reflect the dynamic interaction patterns between brain regions.

[0129] When the dynamic functional connectivity matrix is ​​reduced in dimensionality through nonnegative matrix decomposition, the static network components... Preserving the foundation of time-stable connection patterns, the feature matrix of time-varying brain networks This captures network state fluctuations that evolve over time. Preferably, alternating least squares optimization is used during the decomposition process, and constraints ensure that the decomposition results have clear neurobiological interpretability. Static components correspond to intrinsic connectivity patterns under anatomical constraints, while dynamic components reflect task-independent spontaneous transitions in brain states.

[0130] This embodiment maintains a balance between temporal resolution and statistical power through adaptive window length; phase-locked value calculation avoids interference from signal amplitude fluctuations on functional connectivity estimation; and nonnegative matrix factorization achieves effective separation of stable patterns and dynamic changes while reducing dimensionality. These methods collectively ensure that the time-varying brain network feature matrix can capture rapid fluctuations in neural activity while maintaining a sufficient signal-to-noise ratio for subsequent analysis.

[0131] In some embodiments, white matter fiber bundle topological reconstruction is performed on the diffusion tensor imaging structure connectivity matrix to obtain a structure connectivity weight matrix, including:

[0132] The probabilistic fiber tracing algorithm was used to reconstruct the whole-brain white matter fiber pathways from the diffusion tensor imaging structural connectivity matrix, and the initial structural connectivity strength of the brain region was calculated based on the spatial density distribution of fiber bundles.

[0133] The initial structure connectivity matrix is ​​standardized using graph theory measures to obtain the structure connectivity matrix, which is expressed by formula (4), as follows:

[0134] ;

[0135] In formula (4), brain region With brain regions The initial structural connection strength between them The average anisotropy fraction of the fiber bundle. To determine the number of connecting fibers, The average fiber length, brain region With brain regions The distance between their centroids The width of the Gaussian kernel;

[0136] The structural connection matrix is ​​represented by formula (5), which is as follows:

[0137] ;

[0138] In formula (5), For structural connection matrix;

[0139] By performing a diffusion geometry mapping on the structural connectivity matrix, a low-dimensional manifold space is obtained;

[0140] Spectral clustering algorithm is used to identify features in low-dimensional manifold space to obtain modular features, which include intra-module connectivity enhancement coefficients and inter-module connectivity attenuation coefficients.

[0141] The structural connection matrix is ​​weighted based on modular characteristics, and the structural connection weight matrix is ​​obtained by adjusting the intra-module connection enhancement coefficient and inter-module connection attenuation coefficient using two parameters.

[0142] In this embodiment, when the diffusion tensor imaging structure connectivity matrix is ​​processed by the probabilistic fiber tracing algorithm, a multi-fiber orientation distribution model is used to overcome the analytical challenge of crossed fibers. The calculation of the spatial density distribution of fiber bundles is achieved by statistically counting the number of streamlines passing through each brain region pair, while also considering the influence of partial volume effects. (Brain regions) With brain regions Initial structural connection strength between The quantification integrates three biophysical indicators: microstructural integrity, connectivity density, and conduction efficiency. Among them, the Gaussian kernel term... Gaussian kernel width is used to correct for biases in the estimation of connectivity strength caused by the distance between the centroids of brain regions. Automatically calibrated based on the average connection distance of the whole brain.

[0143] Structure connection matrix The standardization process is achieved through diffusion geometry mapping, which projects high-dimensional connection patterns onto a low-dimensional manifold space, preserving key features of the topology. Low-dimensional embedding employs the Laplacian eigenmap algorithm, which, by solving the spectral decomposition problem of the connection matrix, represents brain regions as low-dimensional coordinate points that preserve connection similarity, effectively suppressing noise interference while highlighting the network's organizational principles.

[0144] Preferably, when the spectral clustering algorithm is executed in a low-dimensional manifold space, the intra-module connection enhancement coefficient is determined by calculating the Z-score normalized value of the connection strength within the same cluster, reflecting the density of local connections; the inter-module connection decay coefficient is derived based on the edge betweenness centrality of cross-module connections, characterizing the sparsity trend of long-range connections; during the dual-parameter adjustment process, the intra-module connection enhancement coefficient is nonlinearly amplified using a sigmoid function, while the inter-module connection decay coefficient is suppressed according to an inverse proportional relationship. The final generated structural connection weight matrix retains the original anatomical constraints and enhances the modularity of the network.

[0145] This embodiment improves the reconstruction accuracy of complex fiber pathways through a probabilistic fiber tracing algorithm; diffuse geometric mapping overcomes the sparsity problem of high-dimensional connectivity data; and a two-parameter adjustment mechanism achieves adaptive enhancement of network topology characteristics, ensuring that the structural connectivity weight matrix reflects both the true anatomical connectivity patterns and highlights functionally relevant topological features. The reconstruction process follows the biological characteristics of white matter fibers, and the extraction of modular features provides a structural constraint basis for subsequent multimodal fusion.

[0146] Please see Figure 2 In some embodiments, a probabilistic fiber tracing algorithm is used to reconstruct whole-brain white matter fiber pathways from the diffusion tensor imaging structural connectivity matrix, and the initial structural connectivity strength between brain regions is calculated based on the spatial density distribution of fiber bundles, including:

[0147] S201. In the anisotropic field defined by the structure connection matrix of the diffusion tensor imaging, the propagation path of white matter fibers is simulated by Monte Carlo random walk. The propagation path of white matter fibers in each simulation is obtained by probability sampling based on the direction of the principal eigenvector of the local diffusion tensor.

[0148] S202. Perform spatial clustering on the fiber bundle paths generated by repeated sampling, retain the connected paths that pass through the preset white matter skeleton region, and record them as the initial fiber bundle paths.

[0149] S203. Calculate the cross-distribution density of the initial fiber tract paths between each brain region, perform geometric correction in combination with the length of the initial fiber tract paths, and generate an initial structural connectivity matrix with directional weights.

[0150] In step S201, preferably, the anisotropic field is constructed based on the fractional anisotropic map of diffusion tensor imaging. The Monte Carlo random walk simulation generates candidate paths by repeatedly sampling the principal eigenvector directions of the local diffusion tensor, and records the spatial coordinate sequence of the path in each simulation. This process preferably employs tens of thousands of iterations to ensure coverage of the cross fibers, while simultaneously improving computational efficiency through GPU parallel acceleration.

[0151] In step S202, the initial fiber bundle path selection can be achieved through anatomical constraints: the preset white matter skeleton region is delineated with reference to the probability template of the JHU white matter map, spatial clustering uses the density peak algorithm to automatically identify the main pathways, and the retained connected paths must meet the requirement that the spatial angle with the known fiber bundle direction does not exceed 15 degrees, which eliminates noise paths and ensures that the reconstruction results conform to the neuroanatomical rules.

[0152] In step S203, during the generation of the initial structural connection matrix, preferably, the cross distribution density is calculated using the kernel density estimation method, and the length correction factor is dynamically adjusted based on the ratio of the fiber bundle travel distance to the Euclidean distance of the brain region centroid; the introduction of directional weights is achieved by analyzing the angle between the initial fiber bundle path and the normal vector of the brain region surface, and the weight allocation of forward and reverse connections follows the biological laws of axonal projection.

[0153] This embodiment reconstructs whole-brain white matter fiber pathways using a probabilistic fiber tracing algorithm. It simulates fiber propagation paths based on Monte Carlo random walks and selects anatomically reasonable initial fiber tract paths by incorporating spatial constraints of the white matter scaffold region. Finally, it generates an initial structural connectivity matrix with directional weights by correcting for fiber tract intersection density and length. This embodiment improves the reconstruction accuracy of complex fiber intersection regions through a probabilistic sampling mechanism. Spatial clustering and the anatomical constraints of the pre-defined white matter scaffold ensure the biological rationality of the fiber paths. The introduction of directional weights reflects the projection characteristics of neural conduction, providing input features with both spatial accuracy and physiological interpretability for the subsequent generation of the structural connectivity weight matrix, thus supporting the precise quantification of white matter structural connectivity in multimodal brain network analysis.

[0154] In some embodiments, spectral clustering algorithms are used to identify features in a low-dimensional manifold space to obtain modular features, including:

[0155] Calculate multiple eigenvectors based on the Laplacian matrix in a low-dimensional manifold space, and construct an eigenvector space;

[0156] The k-means clustering algorithm is used to partition the feature vector space, resulting in modular partitions of the brain network;

[0157] Calculate the membership degree of each brain region in the modular partition of the brain network to generate a modular feature matrix;

[0158] The structural connection matrix is ​​weighted based on modular characteristics, and adjusted using a dual-parameter method that combines intra-module connection enhancement coefficients and inter-module connection attenuation coefficients to obtain the structural connection weight matrix, which includes:

[0159] An enhancement factor is applied to the connectivity of brain regions belonging to the same module to increase the strength of intra-module connectivity; this factor is denoted as the intra-module connectivity enhancement factor.

[0160] An attenuation coefficient is applied to cross-module brain region connections to suppress the strength of inter-module connections; this coefficient is denoted as the inter-module connection attenuation coefficient.

[0161] Repeat the above steps until all brain regions have been traversed;

[0162] The intra-module connection enhancement coefficients and inter-module connection attenuation coefficients are normalized to generate a structural connection weight matrix with modular topological constraints.

[0163] In this embodiment, the Laplacian matrix of the low-dimensional manifold space is constructed by introducing an adaptive kernel function. This function dynamically adjusts the similarity measure according to the functional connectivity between brain regions, so that the feature vector space can more sensitively capture the functional reorganization features of the brain network.

[0164] Preferably, the initial cluster centers of the k-means clustering algorithm are automatically determined through spectral gap analysis, avoiding the subjectivity of manually setting the number of modules. The generated modular partitions not only reflect the static functional division, but also capture the stable module structure in the dynamic reorganization process of the brain network through iterative optimization.

[0165] In the generation of the modular feature matrix, membership degree calculation employs a probabilistic model based on connection density, comprehensively considering the connection strength and spatial distribution characteristics between brain regions and core nodes of modules. The intra-module connection enhancement coefficient is determined by analyzing the statistical distribution characteristics of intra-module connection strength, using a Gaussian kernel function to selectively enhance strong connections. The inter-module attenuation coefficient is dynamically adjusted based on the topological distance of cross-module connections, applying a stronger attenuation effect to long-range cross-module connections. Normalization uses the median of inter-module connection strength as a benchmark reference value to ensure that the adjusted connection weights maintain a reasonable dynamic range.

[0166] This embodiment combines modular feature extraction with dynamic network analysis. Through an adaptive parameter selection mechanism, the structural connectivity weight matrix reflects the modular reorganization patterns of the brain network under different states. This not only enhances the topological features of functional modules but also preserves network plasticity information through dynamic parameter adjustment, providing a more refined analytical tool for studying disease-related brain network remodeling mechanisms. The final structural connectivity weight matrix conforms to the biological organizational principles of neural networks while adapting to individual differences and state changes, providing more interpretable connectivity features for subsequent multimodal fusion.

[0167] In some embodiments, a four-dimensional correlation tensor is constructed using a neurodynamic model by combining the time-varying brain network feature matrix, the structural connectivity weight matrix, and the T1-weighted anatomical features, including:

[0168] A multimodal feature fusion framework is constructed to map the dynamic phase synchronization mode of the time-varying brain network feature matrix to the energy transfer equation of the neural mass unit;

[0169] A white matter conduction delay constraint is imposed on the structural connectivity weight matrix, and it is coupled to the spatial propagation term of the neural mass unit through a diffusion-reaction equation.

[0170] Cortical thickness and gray matter density distribution were extracted from T1-weighted anatomical features and used as local excitability regulation parameters for neural quality units.

[0171] By integrating the energy transfer equation, spatial propagation term, and excitability regulation parameters through tensor product operations, a four-dimensional correlation tensor with spatiotemporal, connectivity, and morphological characteristics is generated. The spatial dimension is configured as the spatial topological arrangement of neural mass units, the temporal dimension is configured as the oscillation period of dynamic phase synchronization, the connectivity strength dimension is configured as the synaptic weights of white matter conduction, and the gray matter morphological dimension is configured as the changes in cortical thickness gradient.

[0172] In this embodiment, the dynamic phase synchronization mode of the time-varying brain network feature matrix is ​​extracted using Hilbert transform, and its energy transfer equation is constructed using a modified Kuramoto model. This model characterizes the cross-frequency interaction of neural oscillations at different frequency bands by introducing a nonlinear term for phase coupling. The white matter conduction delay constraint of the structural connectivity weight matrix is ​​estimated based on fiber bundle length and conduction velocity from DTI imaging. In the diffusion-response equation, it is represented by a time delay operator of the spatial propagation term. This operator simulates the attenuation effect of the signal in long-range white matter pathways through an exponential decay function.

[0173] Cortical thickness and gray matter density distribution of T1-weighted anatomical features were obtained through surface-based morphological analysis. The cortical thickness gradient was calculated using the Laplacian operator to modulate the local excitability threshold of neural mass units. During the construction of the four-dimensional association tensor, tensor product operations can be performed using the Kronecker product to couple multidimensional features. Preferably, the spatial topological arrangement follows the brain region division of the Desikan-Killiany map. The oscillation period of dynamic phase synchronization is determined by extracting time-frequency features through wavelet transform. The synaptic weight dimension is normalized using the modular features of the structural connectivity matrix.

[0174] This embodiment organically integrates multimodal features through a neurodynamic model. The energy transfer equation captures the dynamic characteristics of functional networks, the diffusion-response equation reflects the conduction constraints of structural connections, and local excitability parameters characterize the influence of anatomical morphology. The four-dimensional correlation tensor not only preserves the original features of each modality but also reveals their nonlinear interactions through tensor operations, providing a unified computational framework for studying the synergistic mechanisms of brain function, structure, and morphology. This method elevates the separate multimodal analysis to a unified four-dimensional representation, enabling a more comprehensive reflection of the brain's multi-scale organizational principles.

[0175] In some embodiments, the dynamic phase synchronization mode of the time-varying brain network feature matrix is ​​mapped to the energy transfer equation of the neural mass unit, which is expressed by formula (6), as follows:

[0176] ;

[0177] In formula (6), The total number of brain regions, Indicates the first The energy state of each neural mass unit This represents the normalized functional connectivity weights of the time-varying brain network feature matrix. Indicates the first The oscillation phase of each neural mass unit Indicates the first The oscillation phase of each neural mass unit This is the energy decay coefficient, which is related to the rate of neurotransmitter metabolism. The intensity of the external input stimulus;

[0178] It is coupled to the spatial propagation term of the neural mass unit through the diffusion-reaction equation, and is expressed by Equation (7), which is as follows:

[0179] ;

[0180] In formula (7), The neuron membrane time constant, For the first The activity potential of a neural mass unit This represents the structural connection weights in the structural connection matrix. For delay time variables, This represents the white matter conduction delay time in brain regions, relative to the average fiber length. Positive correlation It is the Sigmoid activation function. This is a Gaussian noise term;

[0181] By integrating the energy transfer equation, spatial propagation term, and excitability regulation parameter through tensor product operation, a four-dimensional correlation tensor with spatiotemporal-connectivity-morphology four-dimensional characteristics is generated, which is expressed by formula (8), as follows:

[0182] ;

[0183] In formula (8), Represents a four-dimensional correlation tensor. Represents a neural mass unit. Representation of neural mass unit The target neural mass unit of structural connectivity. Represents the time series of energy states. This represents the discretization level of the white matter conduction delay time. Indicates the discretization level The representative value, This represents the tensor product operation. Indicates the first The normalized value of the product of gray matter density and cortical thickness in each brain region.

[0184] In this embodiment, the dynamic phase synchronization mode of the time-varying brain network feature matrix is ​​used to extract instantaneous phase information through Hilbert transform, wherein the sinusoidal term of the phase difference represents the dynamic coupling relationship between brain regions; the energy state of the neural mass unit It reflects the activation level of local neural clusters, and its rate of change is influenced by functional connectivity weights. Modulation, this weight is calculated using a sliding time window to determine the time-varying characteristics of the dynamic functional connection; energy attenuation coefficient. It is related to the metabolic process of GABAergic neurotransmitters, and its value can be determined by fitting population-level neuroelectrophysiological data.

[0185] Neuronal membrane time constant in the diffusion-response equation Determined by cellular electrophysiological properties and correlated with the timescale of synaptic integration in neuronal populations; structural connectivity weights The conduction delay time was obtained using a deterministic algorithm for white matter fiber tracking. With fiber bundle length The conversion relationship is established based on a piecewise linear model of the degree of myelination; the sigmoid activation function simulates the input-output nonlinear characteristics of a neuronal population, and its threshold parameter is calibrated based on local field potential recording data.

[0186] Four-dimensional correlation tensor During the construction process, energy state time series White matter conduction delay time was obtained through sliding time window sampling and was first discretized into discrete levels according to the whole-brain histogram. Each discretization level Corresponding to a representative value , Use representative delay values ​​for each level. Calculations are performed; the normalized value of the product of gray matter density and cortex thickness is calculated. The voxel data is mapped to the standard brain space by a surface registration algorithm, and the parameter reflects the structural integrity of the local microcircuits. The tensor product operation uses high-order singular value decomposition to achieve dimensionality reduction and fusion of the feature space.

[0187] This embodiment integrates time-varying functional connectivity, white matter structural networks, and anatomical features into a unified four-dimensional representation by establishing a coupling mechanism between energy dynamics and structural conduction. Equation (6) describes the dynamic synchronization characteristics of the functional network, Equation (7) characterizes the signal propagation law under structural constraints, and Equation (8) realizes tensor fusion of multimodal features. This method provides a computable modeling framework for studying multi-scale interactions of brain networks through rigorous mathematical formalization, and realizes the organic unity of dynamic functional activities and static structural constraints.

[0188] In some embodiments, a multi-task joint learning of a four-dimensional correlation tensor is performed based on a time-varying graphical neural network model to output a quantitative diagnostic result containing disease classification probability, neural circuit abnormality markers, and cognitive decline trajectory, including:

[0189] A biophysically constrained coupled oscillator module is constructed in a time-varying graphical neural network model, including:

[0190] The features contained in the time dimension of the four-dimensional correlation tensor are mapped to the oscillator phase dynamics equation. The natural frequency of each oscillator unit is determined by the resting-state EEG feature spectrum. The coupling strength is dynamically adjusted by the connection strength dimension in the four-dimensional correlation tensor. The oscillator phase dynamics equation is expressed by formula (9), which is as follows:

[0191] ;

[0192] In formula (9), Indicates the first The instantaneous phase of each oscillator unit, Indicates the first The natural frequency of each oscillator unit, Represents nodes There exists a set of neighboring nodes connected by white matter. This represents a slice of the connection strength dimension of a four-dimensional correlation tensor. Represents the phase coupling function. This is a Gaussian white noise term;

[0193] The instantaneous phase synchronization matrix of the oscillator cluster is calculated using a phase-locked detection algorithm and used as the dynamic connection feature input of the graph neural network.

[0194] Spatiotemporal graph convolution based on anatomical constraints processes dynamic connectivity features. Spatiotemporal graph convolution includes spatial dimension convolution processing and temporal dimension convolution processing. Spatial dimension convolution processing includes spatial graph convolution, and the adjacency matrix of spatial graph convolution is limited to the white matter connectivity path determined by DTI fiber tracing. Temporal dimension convolution processing includes deformable convolution kernels to capture multi-scale features from event-related potentials to functional state fluctuations.

[0195] In the convolution operation of each spatiotemporal graph convolution, the node features and connection weights are updated synchronously until the processing is complete;

[0196] Quantitative diagnostic results are generated through a multi-task attention mechanism, including:

[0197] Multitasking attention mechanisms include disease classification branches, neural circuit abnormality monitoring branches, and cognitive prediction branches;

[0198] The disease classification branch calculates the contribution weight of each brain region to the classification decision through dynamic graph pooling, and obtains the first quantitative result containing the first attention weight;

[0199] The neural circuit anomaly detection branch identifies abnormal pathways where the functional connectivity strength and structural connectivity density deviate significantly through side attention scores, and obtains a second quantification result containing a second attention weight.

[0200] The cognitive prediction branch uses a time-attention algorithm to calculate the fluctuation cycle and obtains a third quantization result containing a third attention weight;

[0201] A quantitative diagnostic result is generated based on the first, second, and third quantitative results.

[0202] In this embodiment, the construction of the time-varying graph neural network model organically combines the multi-scale dynamic characteristics of neural activity with anatomical constraints. The time dimension features of the four-dimensional correlation tensor are mapped in a biophysically interpretable manner through the oscillator phase dynamics equation. The intrinsic frequency of each oscillator unit is determined by the resting-state EEG feature spectrum. Specifically, the power spectral density of the preprocessed EEG signal is estimated, and the dominant frequencies of each frequency band are extracted as the reference oscillation frequencies of the corresponding anatomical brain regions. Then, the macroscopic EEG features are assigned to the microscopic oscillator units through cortical surface registration technology. The phase coupling function must satisfy the periodic boundary conditions, and a modified sine function containing an exponentially decaying term is preferred to simulate the nonlinear coupling characteristics between neural clusters.

[0203] The implementation of the phase-locked detection algorithm needs to consider the non-steady-state characteristics of neural oscillations. A sliding time window is used to calculate the instantaneous phase difference after the Hilbert transform, and the window length is adaptively adjusted according to the target frequency band. When the generated instantaneous phase synchronization matrix is ​​used as the input of dynamic connectivity features, topological filtering of white matter connectivity paths is required, that is, only the phase synchronization relationship within the anatomical connectivity range determined by DTI fiber tracing is retained. This hard constraint can effectively eliminate pseudo-connections caused by volume conduction effects. The spatial dimension processing in spatiotemporal graph convolution strictly follows the anatomical adjacency relationship. The positions of the non-zero elements of its adjacency matrix are determined by the deterministic fiber tracing results, and the weight initialization values ​​are taken from the fractional anisotropy index of diffusion tensor imaging.

[0204] Deformable convolutional kernels exhibit multi-scale characteristics in their temporal operations, with their receptive fields dynamically adjusting based on the local temporal features of the input signal. For transient responses such as event-related potentials, the kernel automatically shrinks to capture changes; while for slowly varying processes such as functional state fluctuations, the kernel expands its time scale. This adaptive mechanism is achieved through learnable positional offsets, the adjustment range of which is limited by physiological conduction delays. The synchronous update mechanism of node features and connection weights is reflected in the fact that after each convolutional operation, not only is the state representation of the oscillator unit updated, but the structure-function coupling coefficient is also corrected based on the gradient information output by the current layer. This bidirectional optimization enables the model to adapt to changes in neural plasticity.

[0205] In the multi-task attention mechanism, the dynamic graph pooling operation of the disease classification branch essentially simulates the brain region importance assessment process in clinical diagnosis. The first attention weight reflects the contribution of abnormal activity in a specific brain region to the discrimination of disease classification. The side attention scoring mechanism used in the neural circuit abnormality detection branch identifies abnormal neural pathways whose functional activities exceed the anatomical connectivity constraints by calculating the Mahalanobis distance between functional connectivity strength and structural connectivity density. These pathways are often associated with pathological compensation or disconnection. The temporal attention algorithm of the cognitive prediction branch focuses on analyzing the periodic characteristics of low-frequency fluctuation signals. The third attention weight corresponds to the importance ranking of fluctuations at different time scales for cognitive function prediction, with particular attention paid to slow oscillation components in the range of 0.01-0.1Hz.

[0206] The generation process of quantitative diagnostic results is essentially a fusion decision-making process of multimodal biomarkers. Preferably, the disease probability output of the first quantitative result is processed using softmax normalization; the abnormal markers of the second quantitative result are characterized by Z-scores to represent the degree of deviation from the normal range; and the cognitive trajectory prediction of the third quantitative result is achieved through time series extrapolation. The final integration of quantitative diagnostic results needs to consider the confidence weighting of each branch, where the weight of the disease classification branch is usually set to 1.5-2 times that of other branches, in order to conform to the decision-making characteristics of clinical diagnosis, which is centered on classification.

[0207] This embodiment constructs a biophysically constrained coupled oscillator module to map the dynamic features of a four-dimensional correlation tensor into a phase dynamic process with neurophysiological significance. Combined with spatiotemporal graph convolution processing constrained by anatomical structure, it achieves an organic fusion of the multi-scale dynamic characteristics of neural activity and white matter connectivity topology. Its beneficial effects are: the dual guarantee of oscillator phase synchronization mechanism and rigid anatomical connectivity constraints effectively improves the interpretability of functional connectivity features; the use of deformable convolution kernels to adaptively capture cross-scale features from event-related potentials to functional state fluctuations enhances the physiological rationality of temporal dynamic modeling; the multi-task attention mechanism generates quantitative diagnostic results based on a strict structure-function coupling relationship, enabling disease classification, abnormal loop detection, and cognitive prediction to form a mutually verifying closed-loop system, ultimately outputting diagnostic results with both biophysical interpretability and clinical decision support value.

[0208] In some embodiments, multimodal brain imaging information and cognitive behavioral information of the user are collected. The multimodal brain imaging information includes resting-state functional magnetic resonance imaging time-series signals, diffusion tensor imaging structural connectivity matrices, and T1-weighted anatomical features, including:

[0209] The resting-state functional magnetic resonance time-series signal was configured to be acquired using a preset functional magnetic resonance imaging sequence;

[0210] The diffusion tensor imaging structure connection matrix is ​​configured to acquire data via a preset diffusion-weighted imaging protocol;

[0211] T1-weighted anatomical features were configured to be acquired using high-resolution three-dimensional structural imaging sequences, and the brain was divided into a preset number of automatically anatomically labeled brain regions using an atlas-based segmentation method.

[0212] In this embodiment, the acquisition of multimodal brain imaging data achieves coordinated characterization through a standardized process of three types of imaging technologies. Preferably, the resting-state functional magnetic resonance imaging time-series signal is acquired using a clinically routine gradient echo planar imaging sequence, with its repetition time parameter set according to typical low-frequency neural oscillation characteristics. The raw signal is preprocessed by head movement correction and frequency band filtering to form a time-series matrix. The diffusion tensor imaging structural connectivity matrix is ​​acquired using a multi-directional diffusion-weighted imaging protocol. The raw diffusion data is processed by tensor reconstruction and deterministic fiber tracing to generate a weighted adjacency matrix reflecting the density of white matter pathways. The T1-weighted anatomical features are acquired using a high-resolution three-dimensional structural imaging sequence. The gray matter volume features are mapped to a preset anatomical template space through a standardized cortical segmentation atlas to form a structural feature vector.

[0213] Cognitive behavioral assessments were conducted synchronously with imaging acquisition, using a combination of standardized neuropsychological scales validated for reliability and validity, covering core cognitive domains such as memory, executive function, and information processing speed. Memory assessment employed a standardized delayed recall paradigm, executive function testing included standard cognitive flexibility measurements, and information processing speed was quantified through a symbol-coded basic task. These behavioral indicators formed a multi-dimensional mapping relationship with neuroimaging features.

[0214] In this embodiment, the functional magnetic resonance imaging (fMRI) time-series signals are registered to the individual anatomical space through rigid body transformation, and then nonlinearly normalized and mapped to a standard brain template; diffusion tensor imaging data are aligned across subjects through white matter scaffold extraction; and cognitive behavioral data are converted to Z-scores for normalization. This processing method ensures that the structural connectivity matrix can constrain the construction range of the functional network, so that the final generated fusion features simultaneously retain the individual anatomical structural characteristics and functional dynamic patterns, providing neuroanatomically interpretable input features for subsequent modeling.

[0215] Please see Figure 3 In a second aspect, this embodiment also provides a brain disease classification system 1, applicable to the brain disease classification method described in the first aspect. The system includes a data acquisition module 11, a data processing module 12, a logic operation module 13, and a result output module 14. The data acquisition module 11 is used to collect multimodal brain imaging information and cognitive behavioral information of users. The multimodal brain imaging information includes resting-state functional magnetic resonance imaging time-series signals, diffusion tensor imaging structural connectivity matrices, and T1-weighted anatomical features. The cognitive behavioral information is obtained through standardized neuropsychological scales. The data processing module 12 is used to perform dynamic functional connectivity analysis on the resting-state functional magnetic resonance imaging time-series signals to obtain a time-varying brain network feature matrix. The dynamic functional connectivity analysis is configured to perform phase synchronization calculation based on a sliding window, and to perform white matter fiber tract topology analysis on the diffusion tensor imaging structural connectivity matrix. The reconstruction yields a structural connection weight matrix. Logical operation module 13 is used to construct a four-dimensional correlation tensor using a neurodynamic model, combining the time-varying brain network feature matrix, structural connection weight matrix, and T1-weighted anatomical features. The four-dimensional correlation tensor has spatial, temporal, connection strength, and gray matter morphology dimensions. Multi-task joint learning is performed on the four-dimensional correlation tensor based on a time-varying graphical neural network model, outputting quantitative diagnostic results including disease classification probabilities, neural circuit abnormality markers, and cognitive decline trajectories. The time-varying graphical neural network model is configured as a biophysically constrained coupled oscillator embedding module. Result output module 14 generates an auxiliary clinical classification report containing individualized brain network remodeling targets, disease progression risk stratification, and treatment response prediction based on the cross-modal alignment of the quantitative diagnostic results and cognitive behavioral information.

[0216] In this embodiment, the brain disease classification system 1 achieves automated analysis of multimodal data through a modular architecture. The data acquisition module 11 directly interfaces with medical imaging equipment to ensure standardized raw data collection; the data processing module 12 employs a parallel computing architecture to simultaneously perform dynamic functional connectivity analysis and white matter topology reconstruction; the logic operation module 13 is equipped with a dedicated tensor operation unit, supporting efficient construction of four-dimensional correlation tensors and distributed training of time-varying graph neural networks; the result output module 14 integrates a clinical knowledge base interface to enable the integration of quantitative diagnostic results with the medical decision-making system. All modules communicate via a standardized data bus, ensuring full traceability from raw data input to clinical report generation.

[0217] The system in this embodiment achieves efficient processing of multimodal big data through hardware acceleration, and the interpretability design of the neurodynamic model ensures the reliability of clinical decision-making. Its modular architecture allows for flexible adaptation to the imaging equipment and diagnostic needs of different medical institutions. The resulting auxiliary clinical classification report not only includes disease classification results but also provides personalized treatment recommendations based on neurobiological evidence, enabling rapid translation from scientific research findings to clinical applications.

[0218] By adopting the above technical solutions, this invention differs from existing technologies and possesses the following beneficial effects: Through the collaborative analysis of multimodal brain imaging information and cognitive behavioral information, high accuracy and interpretability in brain disease classification are achieved. Dynamic functional connectivity analysis of resting-state functional magnetic resonance imaging (fMRI) time-series signals captures the time-varying characteristics of brain networks; topological reconstruction of white matter fiber tracts in diffusion tensor imaging provides anatomical constraints; and T1-weighted anatomical features supplement morphological information. These three elements are organically integrated into a four-dimensional correlation tensor constructed through a neurodynamic model, achieving spatiotemporal-connectivity-morphological feature fusion. A time-varying graphical neural network model, based on a biophysically constrained coupled oscillator embedding module, transforms multimodal features into physiologically meaningful quantitative diagnostic results, including disease classification probabilities, neural circuit abnormality markers, and cognitive function decline trajectories. The final generated auxiliary clinical classification report, through cross-modal alignment, provides individualized brain network remodeling targets, disease progression risk stratification, and treatment response prediction, offering a systematic assessment basis for clinical decision-making, ranging from microscopic neural circuit abnormalities to macroscopic cognitive behavioral changes. This invention overcomes the limitations of single-modality analysis by using multi-scale feature fusion and biophysical constraint modeling to enhance the neurobiological interpretability of diagnostic results while maintaining clinical applicability. It is particularly suitable for the auxiliary diagnosis and classification of neurodegenerative diseases such as Alzheimer's disease and Parkinson's disease.

[0219] Finally, it should be noted that although the above embodiments have been described in the text and drawings of this application, this should not limit the scope of patent protection of this application. Any technical solutions that are based on the essential concept of this application and utilize the content described in the text and drawings of this application, resulting in equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this application.

Claims

1. A brain disease classification method characterized by, Comprise: Collecting multi-modal brain imaging information and cognitive behavior information of a user, the multi-modal brain imaging information comprising resting-state functional magnetic resonance time series signals, diffusion tensor imaging structural connection matrices and T1-weighted anatomical features, and the cognitive behavior information being obtained through standardized neuropsychological scales; Performing dynamic functional connection analysis on the resting-state functional magnetic resonance time series signals to obtain a time-varying brain network feature matrix, the dynamic functional connection analysis being configured as phase synchronization calculation based on a sliding window, and performing white matter fiber bundle topological reconstruction on the diffusion tensor imaging structural connection matrices to obtain a structural connection weight matrix; Constructing a four-dimensional correlation tensor through a neural dynamics model based on the time-varying brain network feature matrix, the structural connection weight matrix and the T1-weighted anatomical features, the dimensions of the four-dimensional correlation tensor comprising a spatial dimension, a time dimension, a connection strength dimension and a gray matter morphology dimension; Performing multi-task joint learning on the four-dimensional correlation tensor based on a time-varying graph neural network model to output quantitative diagnostic results containing disease classification probability, neural circuit abnormality markers and cognitive function decline trajectories, the time-varying graph neural network model being configured as a coupled oscillator embedding module with biophysical constraints; Generating an auxiliary clinical classification report containing individualized brain network remodeling targets, disease progression risk stratification and treatment response prediction according to the cross-modal alignment degree of the quantitative diagnostic results and the cognitive behavior information; Wherein, the dynamic functional connection analysis on the resting-state functional magnetic resonance time series signals to obtain a time-varying brain network feature matrix, the dynamic functional connection analysis being configured as phase synchronization calculation based on a sliding window, comprises: Performing sliding window analysis on the resting-state functional magnetic resonance time series signals to obtain a plurality of sliding windows, the sliding window analysis being configured as sliding window algorithm operation based on an adaptive window length, and the window length of the sliding window being dynamically adjusted according to the signal-to-noise ratio of a plurality of brain region signals; Calculating the phase synchronization strength of the brain region corresponding to each sliding window, which is represented by formula (1), and the formula (1) is as follows: ； In formula (1), is the time point, the phase synchrony strength between the brain region , is the total number of time points in a single sliding window, is the brain region , is the instantaneous phase of the time point, is the brain region , is the instantaneous phase of the time point, is the complex exponential function for calculating the phase difference vector between the brain region and the brain region , is the complex number, ; Repeating the above steps until the phase synchronization strength of all brain regions is calculated, and constructing a two-dimensional dynamic functional connection matrix according to the phase synchronization strength, which is represented by formula (2), and the formula (2) is as follows: ； In equation (2), is the dynamic functional connectivity matrix; Performing dimension reduction on the dynamic functional connection matrix through non-negative matrix factorization to extract a time-varying brain network feature matrix, which is represented by formula (3), and the formula (3) is as follows: ； In equation (3), is the static network component after non-negative matrix factorization, is the time-varying brain network feature matrix.

2. The brain disease classification method according to claim 1, characterized by, Performing white matter fiber bundle topological reconstruction on the diffusion tensor imaging structural connection matrices to obtain a structural connection weight matrix, comprising: Reconstructing whole-brain white matter fiber pathways using a probabilistic fiber tracking algorithm on the diffusion tensor imaging structural connection matrices, and calculating the initial structural connection strength of the brain region based on the spatial density distribution of the fiber bundle; Performing standardization processing on the initial structural connection matrix through graph theory measures to obtain a structural connection matrix, which is represented by formula (4), and the formula (4) is as follows: ； In equation (4), is the initial structural connection strength between the brain regions is the initial structural connection strength between the brain regions is the initial structural connection strength between the brain regions is the average fiber bundle anisotropy fraction, is the number of connecting fibers, is the average fiber length, is the initial structural connection strength between the brain regions is the initial structural connection strength between the brain regions is the initial structural connection strength between the brain regions is the Gaussian kernel width; The structural connection matrix is represented by formula (5), and the formula (5) is as follows: ； In equation (5), is the structure connection matrix; Performing diffusion geometry mapping on the structural connection matrix to obtain a low-dimensional manifold space; perform feature recognition on the low-dimensional manifold space by using a spectral clustering algorithm to obtain modular features, the modular features including an intra-module connection enhancement coefficient and an inter-module connection attenuation coefficient; weight the structural connection matrix according to the modular features, and perform two-parameter adjustment on the intra-module connection enhancement coefficient and the inter-module connection attenuation coefficient to obtain a structural connection weight matrix.

3. The brain disease classification method according to claim 2, characterized by, reconstruct whole-brain white matter fiber pathways by using a probabilistic fiber tracking algorithm on the diffusion tensor imaging structural connection matrix, and calculate initial structural connection strengths between brain regions based on spatial density distribution of fiber bundles, including: in an anisotropic field defined by the diffusion tensor imaging structural connection matrix, simulate white matter fiber propagation paths by using a Monte Carlo random walk, and perform probabilistic sampling on a propagation path of each white matter fiber based on a local diffusion tensor principal eigenvector direction to obtain a plurality of fiber bundle paths; perform spatial clustering on the fiber bundle paths generated by repeated sampling, and retain connected paths passing through a preset white matter skeleton region as initial fiber bundle paths; statistically count cross-distribution density of the initial fiber bundle paths between each brain region, and perform geometric correction on the initial fiber bundle paths in combination with lengths of the initial fiber bundle paths to generate an initial structural connection matrix with directional weights.

4. The brain disease classification method according to claim 2, characterized by, perform feature recognition on the low-dimensional manifold space by using a spectral clustering algorithm to obtain modular features, including: calculate a plurality of eigenvectors based on a Laplacian matrix of the low-dimensional manifold space, and construct an eigenvector space; divide the eigenvector space by using a k-means clustering algorithm to obtain modular partitions of a brain network; calculate membership degrees of each brain region in a module in the modular partitions of the brain network to generate a modular feature matrix; weight the structural connection matrix according to the modular features, and perform two-parameter adjustment on the intra-module connection enhancement coefficient and the inter-module connection attenuation coefficient to obtain a structural connection weight matrix, including: apply an enhancement coefficient to a connection between brain regions belonging to the same module to improve intra-module connection strength, and record the enhancement coefficient as an intra-module connection enhancement coefficient; apply an attenuation coefficient to a connection between brain regions across modules to suppress inter-module connection strength, and record the attenuation coefficient as an inter-module connection attenuation coefficient; repeat the above steps until all brain regions are traversed; perform normalization processing on the plurality of intra-module connection enhancement coefficients and inter-module connection attenuation coefficients to generate a structural connection weight matrix with modular topological constraints.

5. The brain disease classification method according to claim 1, characterized by, construct a four-dimensional correlation tensor by using a neural dynamics model on the time-varying brain network feature matrix, the structural connection weight matrix, and T1-weighted anatomical features, including: construct a multi-modal feature fusion framework to map a dynamic phase synchronization mode of the time-varying brain network feature matrix to an energy transfer equation of a neural mass unit; apply a white matter conduction delay constraint to the structural connection weight matrix, and couple the structural connection weight matrix to a spatial propagation term of the neural mass unit by using a diffusion-reaction equation; extract a cortical thickness and a gray matter density distribution in the T1-weighted anatomical features as a local excitability adjustment parameter of the neural mass unit; The energy transfer equation, the spatial propagation term, and the excitability regulation parameter are integrated by a tensor product operation to generate a four-dimensional correlation tensor with spatiotemporal-connection-morphology four-dimensional characteristics, the spatial dimension is configured as the spatial topological arrangement of the neural mass unit, the time dimension is configured as the oscillation period of dynamic phase synchronization, the connection strength dimension is configured as the synaptic weight of white matter conduction, and the gray matter morphology dimension is configured as the cortical thickness gradient change.

6. The brain disease classification method according to claim 5, characterized by, The dynamic phase synchronization mode of the time-varying brain network feature matrix is mapped to the energy transfer equation of the neural mass unit, which is represented by formula (6) as follows: ； In formula (6), is the total number of brain regions, denotes the energy state of the th neural mass unit, denotes the functional connection weight of the normalized time-varying brain network feature matrix, denotes the oscillation phase of the th neural mass unit, denotes the oscillation phase of the th neural mass unit, is the energy decay coefficient, is the external input stimulus intensity; The spatial propagation term is coupled to the neural mass unit by a diffusion-reaction equation, which is represented by formula (7) as follows: ； In Equation (7), is the membrane time constant of the neuron, is the action potential of the th neuron mass unit, denotes the structural connection weight in the structural connection matrix, is the delay time variable, denotes the white matter conduction delay time between brain regions, is the Sigmoid activation function, is the Gaussian noise term; The energy transfer equation, the spatial propagation term, and the excitability regulation parameter are integrated by a tensor product operation to generate a four-dimensional correlation tensor with spatiotemporal-connection-morphology four-dimensional characteristics, which is represented by formula (8) as follows: ； In equation (8), denotes the four-dimensional correlation tensor, denotes the neural mass unit, denotes the neural mass unit connected to the target neural mass unit, denotes the energy state time series, denotes the discretized levels of white matter conduction delay time, denotes the discretized levels of representative values, denotes the tensor product operation, denotes the product normalized value of the gray matter density and the cortical thickness of the brain region.

7. The brain disease classification method according to claim 1, characterized by, The four-dimensional correlation tensor is subjected to multi-task joint learning based on a time-varying graph neural network model, and quantitative diagnostic results containing disease classification probability, neural circuit abnormality marker, and cognitive function decline trajectory are output, including: A biophysical constraint coupled oscillator module is constructed in the time-varying graph neural network model, including: The features contained in the time dimension of the four-dimensional correlation tensor are mapped to the oscillator phase dynamics equation, the intrinsic frequency of each oscillator unit is determined by the resting state electroencephalogram feature spectrum, and the coupling strength is dynamically adjusted by the connection strength dimension in the four-dimensional correlation tensor; A phase-locked detection algorithm is used to calculate the instantaneous phase synchronization matrix of the oscillator cluster as the dynamic connection feature input of the graph neural network; The dynamic connection features are processed based on an anatomical structure constraint spatiotemporal graph convolution, which includes spatial dimension convolution processing and time dimension convolution processing, the spatial dimension convolution processing includes spatial graph convolution, the adjacency matrix of the spatial graph convolution is limited within the white matter connection path range determined by DTI fiber tracking, and the time dimension convolution processing includes a deformable convolution kernel that captures multi-scale features from event-related potentials to functional state fluctuations; The node features and connection weights are synchronously updated in the convolution operation of each layer of spatiotemporal graph convolution until the processing is completed; Quantitative diagnostic results are generated by a multi-task attention mechanism, including: The multi-task attention mechanism includes a disease classification branch, a neural circuit abnormality monitoring branch, and a cognitive prediction branch; The disease classification branch calculates the contribution weight of each brain region to the classification decision by dynamic graph pooling to obtain a first quantitative result containing a first attention weight; The neural circuit abnormality detection branch identifies abnormal pathways with significant deviation of functional connection strength and structural connection density by edge attention score to obtain a second quantitative result containing a second attention weight; The cognitive prediction branch calculates the fluctuation period by a time attention algorithm to obtain a third quantitative result containing a third attention weight; The quantitative diagnostic results are generated according to the first quantitative result, the second quantitative result, and the third quantitative result.

8. The brain disease classification method according to claim 1, characterized by, Collecting multi-modal brain image information and cognitive behavior information of a user, the multi-modal brain image information comprising resting-state functional magnetic resonance time series signals, diffusion tensor imaging structural connectivity matrices and T1-weighted anatomical features, comprising: The resting-state functional magnetic resonance time series signals are configured to be acquired using a preset functional magnetic resonance imaging sequence; The diffusion tensor imaging structural connectivity matrices are configured to be collected by a preset diffusion weighted imaging protocol; The T1-weighted anatomical features are configured to be acquired using a high-resolution three-dimensional structural imaging sequence, and the brain is divided into a preset number of automatic anatomical marker brain regions by a graph-based segmentation method.

9. A brain disease classification system characterized by, The system is suitable for the brain disease classification method of any one of claims 1 to 8, comprising: A data acquisition module for collecting multi-modal brain image information and cognitive behavior information of a user, the multi-modal brain image information comprising resting-state functional magnetic resonance time series signals, diffusion tensor imaging structural connectivity matrices and T1-weighted anatomical features, and the cognitive behavior information being obtained by a standardized neuropsychological scale; A data processing module for performing dynamic functional connectivity analysis on the resting-state functional magnetic resonance time series signals to obtain a time-varying brain network feature matrix, the dynamic functional connectivity analysis being configured to be based on phase synchronization calculation of a sliding window, and performing white matter fiber bundle topology reconstruction on the diffusion tensor imaging structural connectivity matrices to obtain a structural connectivity weight matrix; A logic operation module for constructing a four-dimensional correlation tensor through a neural dynamics model using the time-varying brain network feature matrix, the structural connectivity weight matrix and the T1-weighted anatomical features, the dimensions of the four-dimensional correlation tensor comprising spatial dimensions, time dimensions, connection strength dimensions and gray matter morphology dimensions; performing multi-task joint learning on the four-dimensional correlation tensor based on a time-varying graph neural network model, and outputting a quantitative diagnostic result containing disease classification probability, neural circuit abnormality marker and cognitive function decline trajectory, the time-varying graph neural network model being configured to have a biophysical constraint coupled oscillator embedding module; A result output module for generating an auxiliary clinical classification report containing individualized brain network remodeling targets, disease progression risk stratification and treatment response prediction according to the cross-modal alignment degree of the quantitative diagnostic result and the cognitive behavior information.

Images