Multi-mode brain function connection brain tumor postoperative post-traumatic stress disorder classification system and method
The multimodal brain functional connectivity system solves the problems of objectivity and interpretability in the diagnosis of PTSD after brain tumor surgery, achieves high-precision PTSD classification and early prediction, provides personalized reports of brain network abnormalities, and supports clinical intervention.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GENERAL HOSPITAL OF SOUTHERN THEATRE COMMAND OF PLA
- Filing Date
- 2026-02-15
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies for diagnosing post-traumatic stress disorder (PTSD) after brain tumor surgery suffer from several problems, including a lack of objective biological markers, underutilization of multimodal data, poor interpretability of classification models, and a lack of adaptation to functional atlases specific to brain tumor surgery.
A multimodal brain functional connectivity system is employed, including multimodal data acquisition, multi-map brain network construction, multi-kernel graph convolution feature extraction, and dynamic prediction model construction. Combined with interpretable visualization, a multi-scale brain functional connectivity network is constructed using structural MRI, resting-state and task-state fMRI, EEG, and clinical data to predict the development trajectory of PTSD and provide visual interpretation.
It achieves high-precision PTSD classification and early prediction, with a 9% improvement in classification accuracy. Early data can predict symptom severity up to 14 months later, providing personalized brain network abnormality reports to assist clinical intervention.
Smart Images
Figure CN122087540A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image processing and neuropsychiatric disease diagnosis technology, specifically a classification system and method for post-traumatic stress disorder after brain tumor surgery with multimodal brain functional connectivity. Background Technology
[0002] Brain tumors are a common neurological disease, and their incidence is increasing year by year. Although surgery is the main treatment for brain tumors, postoperative patients often face a variety of psychological challenges, among which post-traumatic stress disorder (PTSD) is a mental complication that seriously affects patients' quality of life. Studies have shown that approximately 15-30% of brain tumor patients will experience PTSD symptoms that meet the diagnostic criteria after surgery, manifested as recurrent replay of traumatic memories, avoidance behavior, negative changes in cognition and emotion, and hypervigilance.
[0003] Currently, the diagnosis of PTSD after brain tumor surgery mainly relies on clinical interviews and scale assessments, such as the Clinically Applied Post-Traumatic Stress Disorder Diagnostic Scale (CAPS-5). Although these methods are widely used, they have significant limitations: First, scale assessments are influenced by patient subjective reports and physician clinical experience, lacking objective biological markers; second, brain tumor patients often have cognitive and language impairments, which may affect the accurate reporting of symptoms; and third, delays in diagnosis after symptom onset lead to missed opportunities for early intervention.
[0004] In recent years, neuroimaging techniques have provided new avenues for the objective diagnosis of mental disorders. fMRI can non-invasively assess brain functional activity using blood oxygen level-dependent (BOLD) signals, while electroencephalography (EEG) can directly record the spatiotemporal dynamics of neural electrical activity. Existing research has attempted to utilize these techniques to explore the neural mechanisms of PTSD, for example:
[0005] Functional connectivity analysis: Zhu et al. used graph neural networks to classify PTSD patients and healthy subjects, achieving an accuracy of 80%. However, this method relies on a single brain atlas to delineate functional connectivity, failing to fully utilize multi-scale brain network information.
[0006] Multimodal fusion approach: A recent study proposed a multimodal Transformer framework that integrates EEG and interview data for depression detection, improving accuracy and precision by 4.7% and 10%, respectively. This indicates that multimodal data fusion can significantly improve the performance of mental disorder classification, but this method was not optimized for the specific population of PTSD after brain tumor surgery.
[0007] Connectome predictive modeling: Ben-Zion et al. developed a connectome-based predictive model (CPM) that can predict the severity of PTSD symptoms 14 months after trauma using fMRI data from one month post-traumatic stress disorder (PTSD). This study confirms the association between large-scale brain networks and the development of PTSD symptoms, but the model's predictive efficacy in the specific population following brain tumor surgery remains unclear.
[0008] It is particularly important to note that postoperative PTSD in brain tumor patients has unique neural mechanisms. The tumor itself and surgical procedures may cause changes in the structure and function of brain networks. These changes, combined with PTSD-related functional abnormalities, result in complex clinical manifestations. Existing neuroimaging studies on PTSD mostly focus on trauma-exposed populations (such as accident and disaster survivors), lacking specialized algorithms for this specific population after brain tumor surgery.
[0009] In addition, existing technologies have the following limitations: (1) they mostly use single-modality image data and fail to fully utilize the complementarity of multimodal brain functional information; (2) the classification models have poor interpretability and are difficult to provide biological markers that clinicians can trust; and (3) they lack functional atlas adaptations specific to brain networks after brain tumor surgery. To address these limitations, this invention provides a classification system and method for post-traumatic stress disorder after brain tumor surgery that addresses multimodal brain functional connectivity. Summary of the Invention
[0010] To address the shortcomings of existing technologies, this invention provides a classification system and method for post-traumatic stress disorder after brain tumor surgery, based on multimodal brain functional connectivity, in order to solve the aforementioned problems.
[0011] To achieve the above objectives, the present invention provides the following technical solution: a multimodal brain functional connectivity classification system for post-traumatic stress disorder following brain tumor surgery, comprising:
[0012] A multimodal data acquisition module is used to acquire multimodal brain function data from patients after brain tumor surgery.
[0013] A multi-map brain network construction module for constructing functional connectivity networks based on brain maps at at least three different scales;
[0014] The multi-kernel graph convolutional feature extraction module is used to extract discriminative features from multi-graph functional connectivity networks;
[0015] A dynamic prediction model building module is used to predict the development trajectory of PTSD based on extracted features;
[0016] The interpretability visualization module is used to provide visual explanations of classification decisions.
[0017] Preferably, the multimodal data acquisition module includes:
[0018] The structural magnetic resonance imaging unit acquires high-resolution T1-weighted structural images for brain region localization and spatial normalization.
[0019] The resting-state functional magnetic resonance imaging unit acquires at least 10 minutes of resting-state BOLD signals for functional connectivity analysis.
[0020] The task-oriented functional magnetic resonance imaging unit employs an emotion-face recognition task to activate fear-related neural circuits.
[0021] The electroencephalography (EEG) unit simultaneously records 64-lead resting-state EEG signals, with a focus on analyzing oscillatory activity in the theta band (4-8 Hz).
[0022] The clinical data integration unit collects tumor-related clinical data and symptom assessment data.
[0023] Preferably, the multi-map brain network construction module uses three brain atlases—AAL116, CC200, and Power264—to divide regions of interest. The AAL116 atlas provides fine-grained division of 116 anatomical brain regions, the CC200 atlas provides medium-scale division of 200 functional brain regions, and the Power264 atlas provides coarse-grained division of 264 functional nodes. Furthermore, it calculates the functional connectivity matrix based on rs-fMRI, the activation synergy matrix based on task-state fMRI, and the coherence matrix based on the EEG theta band.
[0024] Preferably, the multi-kernel graph convolution feature extraction module uses multi-scale convolution kernels to process the graph in parallel and introduces a graph attention mechanism to adaptively weight the importance of different brain nodes.
[0025] Preferably, the dynamic prediction model construction module is based on a connectome prediction modeling framework, which can use early postoperative brain function data to predict the short-term and long-term PTSD symptom development trajectory, and establish specific prediction models for different symptom clusters. The dynamic prediction model construction module specifically includes the following steps:
[0026] Input: Multimodal brain function data collected within one month post-surgery;
[0027] Output: Prediction of PTSD symptom severity in the short term (3-6 months) and long term (12-14 months);
[0028] Symptom cluster-specific prediction: Predictive models were established for the four core symptom clusters of PTSD (intrusive memories, avoidance behavior, negative cognitive and emotional changes, and hypervigilance).
[0029] Preferably, the interpretability visualization module provides class activation mapping visualization, identifies the brain regions and functional connections that contribute most to classification, and generates personalized brain network abnormality reports.
[0030] A classification method for post-traumatic stress disorder after brain tumor surgery using multimodal brain functional connectivity includes the following steps:
[0031] Step 1, Data Acquisition: Collect multimodal brain function data from patients after brain tumor surgery;
[0032] Step 2, Data Preprocessing: Correct, standardize, and filter the collected data to obtain accurate data;
[0033] Step 3: Construct a multi-map brain network: For each modality of data, construct a functional connectivity network based on three brain maps (AAL116, CC200, Power264);
[0034] Step 4, Feature Extraction: Use a multi-kernel graph convolutional network to extract fused features from multi-graph functional connections;
[0035] Step 5: Multi-kernel graph convolution model training: Train the PTSD classification and prediction model based on fused features;
[0036] Step 6, Model Performance Evaluation: Evaluate the model performance on an independent test set;
[0037] Step 7, Interpretability Analysis: Through class activation mapping analysis, the 10 brain regions that contribute the most to the classification of PTSD after brain tumor surgery were identified.
[0038] Preferably, the multi-kernel graph convolutional network training uses a focal loss function to address the class imbalance problem and employs five-fold cross-validation to evaluate model performance.
[0039] Preferably, the brain regions include the medial prefrontal cortex, amygdala, posterior cingulate cortex, anterior insula, dorsolateral prefrontal cortex, hippocampus, anterior cingulate cortex, temporoparietal junction, thalamus, and primary visual cortex.
[0040] Preferably, the interpretability analysis includes visualization of the deviation between the individual's brain network and the normal reference group, longitudinal brain network change trends, and correlation analysis with changes in clinical symptoms.
[0041] Beneficial effects
[0042] Compared with the prior art, the present invention has the following advantages:
[0043] High classification accuracy: By fusing multimodal brain function data and multiscale features, a classification accuracy of 89.3% was achieved on the independent test set, which is about 9% higher than the single-modal method (compared to 84.75% accuracy in Comparison-2).
[0044] Strong early predictive ability: Using brain function data in the early postoperative period (within 1 month), the severity of PTSD symptoms can be predicted 14 months later (predictive correlation ρ=0.31, P<0.001), which is better than the existing CPM model (ρ=0.24).
[0045] Excellent clinical interpretability: Through interpretable AI technology, 10 key brain regions most associated with PTSD after brain tumor surgery were identified, including the medial prefrontal cortex, amygdala, posterior cingulate cortex, and anterior insula, which is consistent with neurobiological research -5.
[0046] Personalized diagnosis: Provides individualized brain network functional abnormality atlases to assist clinicians in developing targeted neuromodulation intervention strategies. Attached Figure Description
[0047] Figure 1 This is a schematic diagram of the system modules of the present invention;
[0048] Figure 2 This is a flowchart of the dynamic prediction model construction module in this invention;
[0049] Figure 3 This is a flowchart of the method of the present invention. Detailed Implementation
[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0051] Please see Figure 1-2 A multimodal brain functional connectivity classification system for post-traumatic stress disorder following brain tumor surgery, comprising:
[0052] A multimodal data acquisition module is used to acquire multimodal brain function data from patients after brain tumor surgery.
[0053] A multi-map brain network construction module for constructing functional connectivity networks based on brain maps at at least three different scales;
[0054] The multi-kernel graph convolutional feature extraction module is used to extract discriminative features from multi-graph functional connectivity networks;
[0055] A dynamic prediction model building module is used to predict the development trajectory of PTSD based on extracted features;
[0056] The interpretability visualization module is used to provide visual explanations of classification decisions.
[0057] Specifically, the multimodal data acquisition module includes:
[0058] The structural magnetic resonance imaging (SMRI) unit acquires high-resolution T1-weighted structural images for brain region localization and spatial normalization; the spatial normalization results of the SMRI unit provide a precise anatomical basis for subsequent brain network construction.
[0059] The resting-state functional magnetic resonance imaging unit acquires at least 10 minutes of resting-state BOLD signals for functional connectivity analysis.
[0060] The task-oriented functional magnetic resonance imaging (fMRI) unit employs an emotion-face recognition task to activate fear-related neural circuits. The BOLD signals from the resting-state and task-oriented fMRI units are combined with brain maps at different scales, and a multi-map brain network construction module generates a functional connectivity network covering both global and local areas.
[0061] The EEG unit simultaneously records 64 leads of resting-state EEG signals, focusing on analyzing the oscillatory activity in the theta band (4-8Hz). The theta band oscillation data from the EEG unit is converted into an EEG functional connectivity matrix, which, together with the fMRI connectivity network, is input into a multi-kernel graph convolutional feature extraction module. This module fuses multimodal features through a multi-kernel learning strategy to extract key feature vectors with significant discriminative power for PTSD classification.
[0062] The clinical data integration unit collects tumor-related clinical data and symptom assessment data.
[0063] With the coordinated efforts of the aforementioned units, the synchronization and integration of multi-dimensional data can be achieved.
[0064] Specifically, the multi-map brain network construction module uses three brain atlases—AAL116, CC200, and Power264—to delineate regions of interest. The AAL116 atlas provides fine-grained delineation of 116 anatomical brain regions, the CC200 atlas provides medium-scale delineation of 200 functional brain regions, and the Power264 atlas provides coarse-grained delineation of 264 functional nodes. It also calculates functional connectivity matrices based on rs-fMRI, activation synergy matrices based on task-state fMRI, and coherence matrices based on the EEG theta band. This allows for a comprehensive capture of the differences in brain functional connectivity at three levels: macroscopic network topology, mesoscopic functional subsystems, and microscopic brain region interactions, generating a multi-scale, multimodal brain functional connectivity network ensemble.
[0065] Specifically, the multi-kernel graph convolution feature extraction module uses multi-scale convolution kernels to process functional connection graphs in parallel and introduces a graph attention mechanism to adaptively weight the importance of different brain nodes; thus, it can accurately capture the topological associations and functional interaction patterns between nodes in brain networks of different scales.
[0066] Specifically, the dynamic prediction model building module is based on the connectome prediction modeling framework, which can use early postoperative brain function data to predict the short-term and long-term PTSD symptom development trajectory, and establish specific prediction models for different symptom clusters. The dynamic prediction model building module specifically includes the following steps:
[0067] Input: Multimodal brain function data collected within one month post-surgery;
[0068] Output: Prediction of PTSD symptom severity in the short term (3-6 months) and long term (12-14 months);
[0069] Symptom cluster-specific prediction: Predictive models were established for the four core symptom clusters of PTSD (intrusive memories, avoidance behavior, negative cognitive and emotional changes, and hypervigilance).
[0070] The above operations can accurately capture the dynamic evolution of PTSD symptoms in patients after brain tumor surgery, significantly improving the accuracy and stability of short-term and long-term symptom prediction. At the same time, specific models for each core symptom cluster can uncover unique brain functional connectivity patterns associated with symptoms such as intrusive memories, avoidance behavior, negative changes in cognition and emotion, and hypervigilance. This provides data support and neural mechanism explanations for the development of targeted intervention plans in clinical practice, helping to achieve early warning and individualized treatment management of PTSD.
[0071] Specifically, the interpretable visualization module provides class activation mapping visualization, identifies the brain regions and functional connections that contribute most to classification, and generates personalized brain network abnormality reports. In other words, the interpretable visualization module can transform complex brain network features into intuitive and easy-to-understand visual charts and text descriptions, helping clinicians quickly locate core brain regions (such as the amygdala, prefrontal cortex, hippocampus, etc.) and functional connectivity pathways that are highly related to the occurrence and development of PTSD.
[0072] Working principle:
[0073] The multimodal data acquisition module completes brain region localization and spatial standardization through the structural MRI unit, and simultaneously acquires BOLD signals from resting / task-state fMRI, 64-lead EEG theta band oscillation data, and tumor clinical and symptom assessment data, achieving simultaneous capture and preliminary integration of multidimensional data;
[0074] The multi-map brain network construction module uses three different scale brain maps, AAL116, CC200, and Power264, to divide regions of interest. It calculates the fMRI functional connectivity matrix, task-state activation synergy matrix, and EEG coherence matrix respectively, generating a multimodal functional connectivity network set covering macroscopic topology, mesoscopic functional subsystems, and microscopic brain region interactions.
[0075] The multi-kernel graph convolution feature extraction module adopts a multi-kernel learning strategy to fuse multi-modal network features and combines graph attention mechanism to adaptively weight key brain nodes to mine feature vectors with significant discriminative power for PTSD classification.
[0076] Then, through the dynamic prediction model construction module, based on the connection group prediction framework, early postoperative data is input, and specific models are built for four core symptom clusters: intrusive memory, avoidance behavior, negative changes in cognition and emotion, and hypervigilance. The prediction results of the severity of PTSD symptoms in the short term (3-6 months) and long term (12-14 months) are output.
[0077] The interpretability visualization module identifies the most contributing brain regions and functional connections through class activation mapping, generating personalized brain network abnormality reports. It transforms complex features into intuitive charts and textual explanations, providing clinicians with a neural mechanism basis for targeted intervention. The entire process achieves full-chain coverage from multimodal data acquisition to clinical decision support, ensuring the accuracy, dynamism, and clinical applicability of the classification system.
[0078] Please see Figure 3 A classification method for post-traumatic stress disorder after brain tumor surgery using multimodal brain functional connectivity includes the following steps:
[0079] Step 1, Data Acquisition: Collect multimodal brain function data from patients after brain tumor surgery;
[0080] Step 2, Data Preprocessing: Correct, standardize, and filter the collected data to obtain accurate data;
[0081] Step 3: Construct a multi-map brain network: For each modality of data, construct a functional connectivity network based on three brain maps (AAL116, CC200, Power264);
[0082] Step 4, Feature Extraction: Use a multi-kernel graph convolutional network to extract fused features from multi-graph functional connections;
[0083] Step 5: Multi-kernel graph convolution model training: Train the PTSD classification and prediction model based on fused features;
[0084] Step 6, Model Performance Evaluation: Evaluate the model performance on an independent test set;
[0085] Step 7, Interpretability Analysis: Through class activation mapping analysis, the 10 brain regions that contribute the most to the classification of PTSD after brain tumor surgery were identified.
[0086] Specifically, the training of the multi-kernel graph convolutional network uses a focal loss function to address the class imbalance problem and employs five-fold cross-validation to evaluate model performance.
[0087] Specifically, the brain regions include the medial prefrontal cortex, amygdala, posterior cingulate cortex, anterior insula, dorsolateral prefrontal cortex, hippocampus, anterior cingulate cortex, temporoparietal junction, thalamus, and primary visual cortex.
[0088] Specifically, interpretability analysis includes visualization of the deviation of individual brain networks from the normal reference group, longitudinal brain network change trends, and correlation analysis with changes in clinical symptoms.
[0089] Example
[0090] Step 1, Data Acquisition: Collect complete multimodal data from 150 patients who underwent brain tumor surgery, including:
[0091] sMRI data: T1-weighted structural images were acquired using a 3T MRI scanner, with a resolution of 1×1×1 mm. 3 ;
[0092] rs-fMRI data: gradient echo planar imaging sequence, TR=2000ms, TE=30ms, voxel size 3×3×3mm 3 Lasting 10 minutes;
[0093] Task-based fMRI: Employing an emotional face recognition task, including fear, neutral, and happy face stimuli;
[0094] EEG data: 64-lead EEG system, sampling rate 1000Hz, bandpass filter 0.5-70Hz, synchronous recording of resting-state EEG for 10 minutes;
[0095] Clinical data: The severity of PTSD symptoms was assessed using CAPS-5 at 1, 3, 6, 12 and 14 months postoperatively.
[0096] Step 2, data preprocessing includes:
[0097] fMRI data: temporal correction, head motion correction, spatial normalization, spatial smoothing, and delinear drift correction;
[0098] EEG data: bad sector removal, filtering, removal of electrooculography artifacts, and noise reduction by independent component analysis;
[0099] Data quality check: Exclude data from subjects with excessive head movement (translation >3mm, rotation >3°) and obvious image artifacts.
[0100] Step 3: Constructing a multi-map brain network: For each modality of data, construct a functional connectivity network based on three brain maps (AAL116, CC200, Power264); specifically including:
[0101] rs-fMRI functional connectivity: Extract the average BOLD time series within each ROI and calculate the Pearson correlation coefficient matrix;
[0102] Task-based fMRI activation synergy: Calculate the activation intensity of each ROI under the contrast of fearful faces vs. neutral faces, and then calculate the activation synergy of brain regions;
[0103] EEG theta band coherence: Extract signals from the 4-8 Hz band, calculate the amplitude squared coherence of each electrode pair, and then project them onto the corresponding brain regions in the source space.
[0104] Step 4, Feature Extraction: Use a multi-kernel graph convolutional network to extract fused features from multi-graph functional connections;
[0105] Step 5: Training the multi-kernel graph convolutional model: This specifically includes...
[0106] The main parameters for constructing a multi-kernel graph convolutional network are as follows:
[0107] Input layer: Receives a 190×190 functional connection matrix (corresponding to 200 ROIs in the CC200 map, which are 190×190 after symmetry processing).
[0108] Multi-kernel graph convolutional layer: Three different sizes of convolutional kernels are used in parallel (hop count = 1, 2, 3), and the output feature dimensions are 64, 128, and 256 respectively;
[0109] Graph Attention Layer: Calculates attention coefficients between nodes and weighted aggregates neighbor node information;
[0110] Feature fusion layer: concatenates multi-scale features and reduces dimensionality through a fully connected layer;
[0111] Output layer: Employs the Sigmoid activation function to output the PTSD probability classification result.
[0112] The model was trained using five-fold cross-validation, with Adam as the optimizer and an initial learning rate of 0.001. Focal loss was used to address the class imbalance problem.
[0113] Step 6, Model Performance Evaluation: Evaluate the model performance on the independent test set. The results show:
[0114] Table 1. Comparison of classification performance of different methods
[0115] Classification methods Accuracy (%) Sensitivity (%) Specificity (%) AUC Single AAL map 82.4 81.6 83.1 0.841 Single CC200 spectrum 84.8 83.9 85.6 0.850 Multi-map fusion (without multi-core) 86.5 85.2 87.7 0.872 This invention (multi-map + multi-core) 89.3 88.7 89.9 0.912
[0116] At the same time, the model also performs excellently in symptom prediction:
[0117] Data from 1 month post-surgery predicted PTSD symptom severity 14 months later: ρ=0.31, P<0.001;
[0118] The highest accuracy was found in predicting invasive memory symptoms (AUC=0.894).
[0119] The accuracy of predicting avoidance behavior symptoms was the second highest (AUC=0.872).
[0120] Step 7, Interpretability Analysis: Through class activation mapping analysis, the 10 brain regions that contributed most to the classification of PTSD after brain tumor surgery were identified:
[0121] Table 2 Key Brain Regions Contributing to PTSD Classification
[0122] Brain region names Network Contribution Medial prefrontal cortex (mPFC) Default network 12.3% Right amygdala Highlighting the network 11.8% PCC (Posterior buckle return) Default network 9.7% Maejima Hajime Highlighting the network 8.9% dorsolateral prefrontal cortex Central Execution Network 7.6% hippocampus Edge systems 7.2% Front buckle back Highlighting the network 6.8% Temporoparietal junction Default network 6.1% thalamus subcutaneous structures 5.9% primary visual cortex Visual Networks 4.7%
[0123] These brain regions are closely related to fear processing, emotion regulation, and self-referential thinking, consistent with neurobiological models of PTSD-3-5. In particular, it was found that in patients with brain tumors, tumor location is closely related to changes in the strength of functional connectivity in these key brain regions; for example, the risk of PTSD is significantly increased when the tumor involves nodes of the default mode network or salience network.
[0124] Clinical validation:
[0125] A prospective clinical validation was conducted on 30 patients who underwent brain tumor surgery. Psychiatrists assessed the patients using both traditional clinical assessment and the system of this invention without their knowledge. Results showed that compared to traditional clinical assessment, the diagnostic results of this invention were more consistent with the independent diagnoses of psychiatrists (Kappa=0.78 vs 0.62); the average diagnostic time was significantly shortened (systematic analysis: 2.5 hours vs comprehensive clinical assessment: 3.5 days); and patients had a higher level of understanding and acceptance of the brain function report (90% of patients considered the report "easy to understand" and "helpful").
[0126] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.
[0127] It should be noted that, in this document, relational 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 such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0128] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A classification system for post-traumatic stress disorder following brain tumor surgery using multimodal brain functional connectivity, characterized in that, include: A multimodal data acquisition module is used to acquire multimodal brain function data from patients after brain tumor surgery. A multi-map brain network construction module for constructing functional connectivity networks based on brain maps at at least three different scales; The multi-kernel graph convolutional feature extraction module is used to extract discriminative features from the multi-graph functional connectivity network. A dynamic prediction model building module is used to predict the development trajectory of PTSD based on extracted features; The interpretability visualization module is used to provide visual explanations of classification decisions.
2. The multimodal brain functional connectivity classification system for post-traumatic stress disorder after brain tumor surgery according to claim 1, characterized in that, The multimodal data acquisition module includes: The structural magnetic resonance imaging unit acquires high-resolution T1-weighted structural images for brain region localization and spatial normalization. The resting-state functional magnetic resonance imaging unit acquires at least 10 minutes of resting-state BOLD signals for functional connectivity analysis. The task-oriented functional magnetic resonance imaging unit employs an emotion-face recognition task to activate fear-related neural circuits. The electroencephalography (EEG) unit simultaneously records 64-lead resting-state EEG signals, with a focus on analyzing oscillatory activity in the theta band (4-8 Hz). The clinical data integration unit collects tumor-related clinical data and symptom assessment data.
3. The multimodal brain functional connectivity classification system for post-traumatic stress disorder after brain tumor surgery according to claim 1, characterized in that, The multi-map brain network construction module uses three brain atlases—AAL116, CC200, and Power264—to delineate regions of interest. The AAL116 atlas provides fine-grained delineation of 116 anatomical brain regions, the CC200 atlas provides medium-scale delineation of 200 functional brain regions, and the Power264 atlas provides coarse-grained delineation of 264 functional nodes. Furthermore, it calculates the functional connectivity matrix based on rs-fMRI, the activation synergy matrix based on task-state fMRI, and the coherence matrix based on the EEG theta band.
4. The multimodal brain functional connectivity classification system for post-traumatic stress disorder after brain tumor surgery according to claim 1, characterized in that, The multi-kernel graph convolution feature extraction module uses multi-scale convolution kernels to process the graph in parallel and introduces a graph attention mechanism to adaptively weight the importance of different brain nodes.
5. A classification system for post-traumatic stress disorder following brain tumor surgery based on multimodal brain functional connectivity as described in claim 1, characterized in that, The dynamic prediction model construction module is based on a connectome prediction modeling framework, which can use early postoperative brain function data to predict the short-term and long-term PTSD symptom development trajectory, and establish specific prediction models for different symptom clusters. The dynamic prediction model construction module specifically includes the following steps: Input: Multimodal brain function data collected within one month post-surgery; Output: Prediction of PTSD symptom severity in the short term (3-6 months) and long term (12-14 months); Symptom cluster-specific prediction: Predictive models were established for the four core symptom clusters of PTSD (intrusive memories, avoidance behavior, negative cognitive and emotional changes, and hypervigilance).
6. A classification system for post-traumatic stress disorder following brain tumor surgery based on multimodal brain functional connectivity as described in claim 1, characterized in that, The interpretability visualization module provides class activation map visualization, identifies brain regions and functional connections that contribute most to classification, and generates personalized reports of brain network anomalies.
7. A classification method for post-traumatic stress disorder after brain tumor surgery using multimodal brain functional connectivity, characterized in that, Includes the following steps: Step 1, Data Acquisition: Collect multimodal brain function data from patients after brain tumor surgery; Step 2, Data Preprocessing: Correct, standardize, and filter the collected data to obtain accurate data; Step 3: Construct a multi-map brain network: For each modality of data, construct a functional connectivity network based on three brain maps (AAL116, CC200, Power264); Step 4, Feature Extraction: Use a multi-kernel graph convolutional network to extract fused features from multi-graph functional connections; Step 5: Multi-kernel graph convolution model training: Train the PTSD classification and prediction model based on fused features; Step 6, Model Performance Evaluation: Evaluate the model performance on an independent test set; Step 7, Interpretability Analysis: Through class activation mapping analysis, the 10 brain regions that contribute the most to the classification of PTSD after brain tumor surgery were identified.
8. The method for classifying post-traumatic stress disorder after brain tumor surgery according to claim 7, characterized in that, The multi-kernel graph convolutional network training uses a focal loss function to address the class imbalance problem and employs five-fold cross-validation to evaluate model performance.
9. A method for classifying post-traumatic stress disorder after brain tumor surgery using multimodal brain functional connectivity as described in claim 7, characterized in that, The brain regions include the medial prefrontal cortex, amygdala, posterior cingulate cortex, anterior insula, dorsolateral prefrontal cortex, hippocampus, anterior cingulate cortex, temporoparietal junction, thalamus, and primary visual cortex.
10. A method for classifying post-traumatic stress disorder after brain tumor surgery using multimodal brain functional connectivity as described in claim 7, characterized in that, The visualization analysis includes visualization of the deviation between the individual's brain network and the normal reference group, longitudinal brain network change trends, and correlation analysis with changes in clinical symptoms.