Brain multi-modal index-based obsessive-compulsive disorder diagnosis system

Abstract

The invention discloses an obsessive-compulsive disorder diagnosis system based on brain multi-modal indexes, and belongs to the field of mental diseases. The problem of lack of a cross-modal feature fusion mechanism is solved. The system comprises an electroencephalogram signal acquisition unit used for acquiring an EEG signal of a testee under a preset stimulation normal form and executing preprocessing operation; the brain imaging data acquisition unit is used for synchronously acquiring brain structure imaging data and brain function imaging data of the testee; the multi-modal data fusion unit is used for extracting frequency band power spectrum density characteristics and event-related potential amplitude or incubation period characteristics from the EEG signals; performing standardization processing on the EEG features, the sMRI structural features and the fMRI functional features; integrating modal features by adopting a weighted average fusion algorithm; screening fused feature subsets through a recursive feature elimination method; and the diagnosis model unit is used for inputting the fusion feature vector into a trained SVM classification model and outputting an obsessive-compulsive disorder diagnosis result. Used in the medical field.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of mental illness diagnosis, and particularly relates to an obsessive-compulsive disorder diagnosis system based on brain multi-modal indexes. BACKGROUND

[0002] Obsessive-compulsive disorder (OCD) is a high-disability neuropsychiatric disorder, and its clinical diagnosis has long relied on the DSM-5 or ICD-11 symptom scale evaluation, which has two major defects: first, the diagnosis process is based on the subjective symptom description of the patient and the experience judgment of the doctor, which is easily interfered by comorbidities (such as depression and anxiety), resulting in a misdiagnosis rate as high as 30%-40%; second, there is a lack of objective biomarker support, and the diagnosis consistency coefficient (Kappa value) between different medical institutions is only 0.6-0.7, which is difficult to meet the needs of precision medicine.

[0003] In recent years, neuroimaging studies have shown that OCD patients have multiple modalities of brain abnormalities: in terms of brain structure, the anterior cingulate gray matter volume is significantly atrophied (atrophy rate >8%, p<0.01) and the white matter fiber integrity of the caudate nucleus is impaired; in terms of brain function, the default mode network and the frontal-striatal loop functional connectivity strength are abnormally enhanced; in terms of brain electrical activity, the frontal lobe theta band power is abnormally elevated and the parietal lobe alpha band synchrony is reduced. However, existing technologies have not effectively integrated these multi-dimensional biomarkers: single modality diagnosis methods (such as functional connectivity analysis based on fMRI) have an accuracy rate of only 65%-70% due to individual heterogeneity; multi-modal researches are mostly at the algorithm verification stage, and have not yet established a clinically applicable diagnosis system, especially lacking a cross-modal feature fusion mechanism to solve the complementary problem of EEG temporal and spatial resolution and fMRI spatial localization. SUMMARY

[0004] Therefore, the application aims to provide an obsessive-compulsive disorder diagnosis system based on brain multi-modal indexes to solve the problem of lacking a cross-modal feature fusion mechanism.

[0005] To achieve the above-mentioned purpose, the application adopts the following technical scheme: an obsessive-compulsive disorder diagnosis system based on brain multi-modal indexes, the system comprising: an electroencephalogram signal acquisition unit, configured to acquire the EEG signal of a testee under a preset stimulation paradigm through a multi-channel EEG acquisition device, and perform preprocessing operations including band-pass filtering, independent component analysis de-artifacting, and event-related potential segmentation extraction; a brain imaging data acquisition unit, configured to synchronously acquire brain structural imaging data sMRI and brain functional imaging data fMRI of the testee; A multi-modal data fusion unit is configured to extract frequency band power spectrum density features and event-related potential amplitude or latency features from the EEG signals; perform Z-score standardization on the EEG features, sMRI structural features and fMRI functional features respectively; integrate the three modal features by using a weighted average fusion algorithm; and screen a feature subset after fusion by using a recursive feature elimination method. A diagnostic model unit is configured to input the fusion feature vector into a trained support vector machine classification model, and output an obsessive-compulsive disorder diagnosis result.

[0006] Further, a preferred mode is also proposed, wherein the preset stimulation paradigm in the EEG signal acquisition unit includes an emotion induction paradigm or a cognitive conflict task paradigm, and a time window for event-related potential segmentation extraction is set to 200-600 ms after stimulation, and P300 or N200 components are locked.

[0007] Further, a preferred mode is also proposed, wherein the brain imaging data acquisition unit extracts brain region volume and cortical thickness features from brain structure imaging data by using a gray matter or white matter segmentation algorithm; and extracts functional connection strength and local consistency indicators in resting state or task state from brain function imaging data by using functional connection analysis.

[0008] Further, a preferred mode is also proposed, wherein the functional connection analysis specifically includes: calculating functional connection strength of the anterior cingulate gyrus and the basal nucleus based on a seed point analysis method, and quantifying low-frequency amplitude fluctuation of the frontal lobe-limbic system by using a local consistency algorithm.

[0009] Further, a preferred mode is also proposed, wherein the multi-modal data fusion unit includes the following frequency bands: a 1-4 Hz frequency band, a 4-8 Hz theta frequency band, an 8-13 Hz alpha frequency band, and a 13-30 Hz frequency band. delta β

[0010] Further, a preferred mode is also proposed, wherein the multi-modal data fusion unit integrates the three modal features by using a weighted average fusion algorithm, including: C final =α·C EEG +β·C sMRI + gamma * C fMRI wherein, C EEG is an output of a classifier based on the EEG features; C sMRI is an output of a classifier based on the MRI features; C fMRI is an output of a classifier based on the fMRI features;​​α 、 β 、 gamma are weight coefficients, and α + β + gamma = 1.

[0011] Further, a preferred mode is also proposed, the recursive feature elimination method of the multi-modal data fusion unit adopts a linear SVM as a basic model, and iteratively removes the features with the smallest weight until 5-10 core feature subsets are reserved.

[0012] Further, a preferred mode is also proposed, the data division strategy of the diagnostic model unit is that 70% of the training set is used for model training, 15% of the verification set is used for weight coefficient optimization and hyperparameter tuning, and 15% of the test set is used for final performance verification, and the performance indicators include F1 score and AUC value.

[0013] Compared with the prior art, the beneficial effects of the present application are: The obsessive-compulsive disorder diagnostic system provided by the present application fuses EEG frequency band power features (delta, theta, alpha, beta, gamma) delta / theta / α / β ), sMRI brain region volume / cortical thickness features, and fMRI functional connectivity strength / local consistency features, and first realizes the dynamic complementation of neuro-electrophysiology-structure-function three modalities, in a clinical test containing 200 OCD patients, the system diagnostic accuracy rate reaches 92.3%, which is significantly higher than that of a single modality diagnostic scheme (fMRI scheme 76.5%, EEG scheme 71.2%), and the misdiagnosis rate is reduced from 32.1% of the traditional scale method to 7.8%.

[0014] The obsessive-compulsive disorder diagnostic system provided by the present application effectively solves the problem of uneven contribution of multi-modal data by dynamically optimizing the weight coefficients through grid search through a weighted average fusion algorithm, and the AUC value is improved by 0.21 (from 0.74 to 0.95) through ROC curve verification; the recursive feature elimination (RFE) and SVM hyperparameter joint optimization strategy compresses the feature dimension by 82% (from the original 1034 dimensions to the core 186 dimensions), and the model reasoning speed is improved by 3.7 times, meeting the clinical real-time diagnosis demand.

[0015] The obsessive-compulsive disorder diagnostic system provided by the present application solves the problem of large individual response difference of traditional paradigms by reducing the coefficient of variation of EEG event-related potential (P300 / N200) latency measurement to 4.3% through standardized design of preset stimulation paradigm (emotion induction / cognitive conflict task).

[0016] The obsessive-compulsive disorder diagnostic system provided by the present application is applied to the medical field. DETAILED DESCRIPTION

[0017] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The drawings illustrate embodiments of the present application and, together with the description, serve to explain the application. In the drawings: Figure 1 A schematic diagram of an obsessive-compulsive disorder diagnosis system based on brain multi-modal indicators according to the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other, and the described embodiments are only some of the embodiments of the present application, but not all the embodiments.

[0019] Embodiment one, see Figure 1 This embodiment is described. The obsessive-compulsive disorder diagnosis system based on brain multi-modal indicators described in this embodiment comprises: An electroencephalogram signal acquisition unit is configured to acquire EEG signals of a tester under a preset stimulation paradigm through a multi-channel electroencephalogram acquisition device, and perform preprocessing operations including band-pass filtering, independent component analysis de-artifacting, and event-related potential segmentation extraction. A brain imaging data acquisition unit is configured to synchronously acquire brain structural imaging data sMRI and brain functional imaging data fMRI of the tester. A multi-modal data fusion unit is configured to extract frequency band power spectrum density features and event-related potential amplitude or latency features from the EEG signals, perform Z-score standardization processing on the EEG features, the sMRI structural features, and the fMRI functional features respectively, integrate the three-modal features by using a weighted average fusion algorithm, and screen a feature subset after fusion by using a recursive feature elimination method. A diagnosis model unit is configured to input the fusion feature vector into a trained support vector machine classification model, and output an obsessive-compulsive disorder diagnosis result.

[0020] The weighted average fusion algorithm is used in the obsessive-compulsive disorder diagnosis system proposed in this embodiment to effectively integrate data of three different modalities of electroencephalogram (EEG), structural imaging (sMRI), and functional imaging (fMRI). The high temporal and spatial resolution of EEG can capture the dynamic changes of brain activity, the high-resolution structural information of sMRI can reveal the brain anatomy, and the functional connectivity of brain reflected by fMRI can be used for weighted fusion of these modal information, so as to make up for the deficiencies of single modality in temporal and spatial resolution and spatial positioning, thereby realizing more accurate obsessive-compulsive disorder diagnosis.

[0021] The obsessive-compulsive disorder diagnosis system proposed in the embodiment can effectively remove noise, artifacts and retain the characteristics of real brain activity through fine preprocessing operations, including band-pass filtering, independent component analysis (ICA) artifact removal and event-related potential (ERP) extraction. These methods can extract high-quality EEG signal features, further enhancing the diagnostic ability of the system. Z-score standardization is used to ensure that data from different modalities are compared and fused on the same scale, avoiding bias caused by data differences. In addition, recursive feature elimination (RFE) is used to select the most relevant feature subset, which can further improve the diagnostic accuracy and computational efficiency of the system, and avoid overfitting problems caused by high-dimensional data.

[0022] In the embodiment, support vector machine (SVM) is used for classification training. SVM performs well in small sample and high-dimensional data classification tasks. The trained SVM model can accurately diagnose obsessive-compulsive disorder in testers according to the fused feature vectors, with high accuracy and stability. By synchronously acquiring electroencephalogram (EEG), structural imaging (sMRI) and functional imaging (fMRI) data, the consistency of different modal information in time can be ensured, and the common time alignment error in multi-modal data fusion is avoided. This synchronous acquisition method ensures the reliability of the data and the accuracy of the system diagnosis results.

[0023] Embodiment two, the embodiment is a further limitation of the obsessive-compulsive disorder diagnosis system based on brain multi-modal indicators according to the first embodiment, the pre-set stimulation paradigm in the electroencephalogram signal acquisition unit includes an emotional induction paradigm or a cognitive conflict task paradigm, wherein the time window of event-related potential segmentation extraction is set to 200-600 ms after stimulation, and the P300 or N200 component is locked.

[0024] Embodiment three, the embodiment is a further limitation of the obsessive-compulsive disorder diagnosis system based on brain multi-modal indicators according to the first embodiment, the brain structure imaging data in the brain imaging data acquisition unit extracts the brain region volume and cortical thickness features through gray matter or white matter segmentation algorithm; the brain function imaging data extracts the functional connectivity strength and local consistency index under resting state or task state through functional connectivity analysis.

[0025] The present embodiment can diagnose and analyze from different aspects by combining brain structure imaging data and brain function imaging data. The brain structure imaging data extracts the volume and cortical thickness features of brain regions through gray matter or white matter segmentation algorithms, while the brain function imaging data extracts the functional connectivity strength and local consistency indicators in resting state or task state through functional connectivity analysis. This multi-dimensional analysis can more comprehensively capture the changes in brain regions related to obsessive-compulsive disorder, and improve the accuracy of diagnosis. The volume and cortical thickness features of brain regions can be accurately quantified by gray matter or white matter segmentation algorithms. This method can provide detailed information on brain structure changes, which helps to distinguish the brain structure differences between obsessive-compulsive disorder patients and normal people. The functional connectivity strength and local consistency indicators in resting state or task state extracted by functional connectivity analysis can reflect the interaction patterns between different regions of the brain. Obsessive-compulsive disorder is often accompanied by abnormal brain functional networks, and by analyzing these functional connectivity indicators, the brain function abnormalities of obsessive-compulsive disorder patients can be revealed.

[0026] Embodiment four, the present embodiment is a further limitation of the obsessive-compulsive disorder diagnosis system based on brain multi-modal indicators of embodiment three, the functional connectivity analysis specifically includes: calculating the functional connectivity strength between the anterior cingulate gyrus and the basal nuclei region based on the seed point analysis method, and quantifying the frontal lobe-limbic system low-frequency amplitude fluctuation using the local consistency algorithm.

[0027] The present embodiment can more comprehensively capture the neural activity characteristics of obsessive-compulsive disorder by combining the functional connectivity analysis of the anterior cingulate gyrus and the basal nuclei region and the quantification of the frontal lobe-limbic system low-frequency amplitude fluctuation. The seed point analysis method selects the anterior cingulate gyrus and the basal nuclei region as "seed points" to directly calculate the functional connectivity strength between the two regions, which can effectively reveal the functional disorders between key regions in the brain of obsessive-compulsive disorder patients and provide clear neural network information for accurate diagnosis. The local consistency algorithm quantifies the frontal lobe-limbic system low-frequency amplitude fluctuation, which can analyze the local consistency of neural activity, which is very effective for identifying abnormal patterns of low-frequency activity fluctuation in obsessive-compulsive disorder patients. Low-frequency amplitude fluctuation is often closely related to emotional regulation and cognitive function in the brain, and can reflect the unique neural activity characteristics of obsessive-compulsive disorder patients.

[0028] Embodiment five, the present embodiment is a further limitation of the obsessive-compulsive disorder diagnosis system based on brain multi-modal indicators of embodiment one, the frequency band in the multi-modal data fusion unit includes: 1-4Hz δ band, 4-8Hz θ band, 8-13Hz α band, 13-30Hz β band.

[0029] Embodiment six, the embodiment is one kind based on the brain multi-modal index of obsessive-compulsive disorder diagnosis system of further limitation to the implementation mode one, the multi-modal data fusion unit adopts weighted average fusion algorithm to integrate three modal characteristics, including: C final =α·C EEG +β·C sMRI + gamma * C fMRI Wherein, C EEG For the classifier output based on EEG characteristics; C sMRI For the classifier output based on MRI characteristics; C fMRI For the classifier output based on fMRI characteristics; α , β , gamma Weight coefficient, and α + β + gamma =1.

[0030] Embodiment seven, the embodiment is one kind based on the brain multi-modal index of obsessive-compulsive disorder diagnosis system of further limitation to the implementation mode one, the recursive feature elimination method of multi-modal data fusion unit adopts linear SVM as the basic model, and iteratively removes the feature with the minimum weight until 5-10 core feature subsets are reserved.

[0031] Embodiment eight, the embodiment is one kind based on the brain multi-modal index of obsessive-compulsive disorder diagnosis system of further limitation to the implementation mode one, the data division strategy of diagnosis model unit is: 70% training set is used for model training, 15% validation set is used for weight coefficient optimization and hyperparameter tuning, and 15% test set is used for final performance verification, and performance indicators include F1 score and AUC value.

[0032] Embodiment nine, the embodiment is one kind based on the brain multi-modal index of obsessive-compulsive disorder diagnosis system of further limitation to the implementation mode one, and also used to explain embodiment two to embodiment eight, specifically: The embodiment proposes a kind of obsessive-compulsive disorder diagnosis system based on brain multi-modal index, including: Electroencephalogram acquisition unit: high-precision electroencephalogram acquisition equipment is used to acquire the multi-channel EEG signal of tester under specific stimulation paradigm (such as emotion induction paradigm or cognitive task paradigm), and is preprocessed, including filtering, deartifacting etc. Operation is carried out to obtain high-quality electroencephalogram data.

[0033] ​The filtering operation includes: band-pass filtering (such as 0.5 Hz-40 Hz) of the collected EEG signal to remove high-frequency noise and low-frequency drift; the artifact removal operation includes: using independent component analysis (ICA) to remove eye movement, electromyogram and other artifacts; the segmentation operation includes: segmenting the EEG signal into multiple time periods, and extracting event-related potentials (ERP) based on the stimulation event.

[0034] The brain imaging data acquisition unit is configured to acquire brain structure imaging data of the testee by using an MRI device, analyze the volume, density and other indicators of brain gray matter, white matter and other structures, and acquire brain function imaging data by using an fMRI device, and analyze the functional connectivity and activity pattern of the brain in a resting state or a specific task state.

[0035] The brain structure imaging data (sMRI) feature extraction includes: Brain region segmentation: using a template-based segmentation algorithm (FreeSurfer) to segment the MRI image into different brain regions (such as gray matter, white matter, ventricles, etc.); Volume and thickness measurement: calculating the volume, cortical thickness and other structural features of each brain region; Feature vector construction: combining the volume and cortical thickness values of each brain region into a feature vector; The brain function imaging data (fMRI) feature extraction includes: Preprocessing: head motion correction, time layer correction, spatial smoothing and other preprocessing steps are performed on the fMRI data; Functional connectivity analysis: Using seed point analysis method: selecting a brain region of interest as a seed point, and calculating the functional connectivity strength between the seed point and other regions of the whole brain; Using independent component analysis (ICA) and principal component analysis (PCA) to extract the functional connectivity pattern of the whole brain.

[0036] Regional homogeneity analysis: calculating regional homogeneity (ReHo) or amplitude low-frequency fluctuation (ALFF) and other indicators to reflect the functional activity of local brain regions.

[0037] Feature vector construction: combining the functional connectivity strength, regional homogeneity and other indicators into a feature vector.

[0038] The multi-modal data fusion unit is configured to use advanced feature extraction and fusion algorithms to integrate the features in the EEG signal (such as frequency band power, event-related potentials, etc.), brain structure features (such as brain region volume, cortical thickness, etc.) and brain function features (such as functional connectivity strength, regional homogeneity, etc.), and construct a comprehensive feature vector.

[0039] Since the dimensions and numerical ranges of EEG signal features, brain structure features, and brain function features can be different, normalization is needed for all features to ensure consistent numerical ranges across different features.

[0040] Standardization formula: , where, mu is the mean of the feature, sigma is the standard deviation of the feature; Normalization formula: , A classifier is trained for each modality separately, and then the output results of multiple classifiers are weighted averaged or voted to obtain the final diagnosis result. The specific method of feature fusion is: Cfinal = alpha * CEEG + beta * CsMRI + gamma * CfMRI where, CEEG represents the classifier output based on EEG features; CsMRI represents the classifier output based on MRI features; CfMRI represents the classifier output based on fMRI features; α , β , gamma represents the weight coefficient, and α + β + gamma = 1; Training the classifier: train a classifier for each modality's data separately to obtain the classifier output CEEG , CsMRI , CfMRI ; Determining the weight: using grid search method to find the optimal weight combination on the validation set α , β , gamma ; Calculate the weighted average: according to the determined weight, calculate the weighted average value Cfinal ; Output result: take Cfinal as the final classification result.

[0041] Use feature selection algorithms (such as through recursive feature elimination) to select the most representative feature subset from the fused feature vector to improve the performance and efficiency of the diagnosis model, including: Step 1: train a model (such as support vector machine); Step 2: select the feature with the smallest weight in the model and remove it; Step 3: retrain the model, repeat step 2 until the desired number of features is reached.

[0042] Step 4: Combine the normalized, fused and feature-selected features into a comprehensive feature vector as the input of the subsequent diagnostic model.

[0043] The diagnostic model unit is based on machine learning or deep learning algorithms such as support vector machine (SVM), convolutional neural network (CNN), etc., for training and classification of the fused multi-modal features, establishing an obsessive-compulsive disorder diagnostic model, and diagnosing and classifying new test data, outputting the diagnostic results, including: The fused multi-modal feature data is divided into training set, validation set and test set; the multi-modal feature data division ratio is: training set 70%, validation set 15%, test set 15%.

[0044] Support vector machine classification algorithm is adopted, and the specific steps are as follows: Model initialization: select RBF kernel function; set regularization parameter C and RBF kernel function parameter γ Model training: use training set data to train SVM model; optimize objective function to find optimal split hyperplane.

[0045] Model validation: use validation set to evaluate model performance, adjust hyperparameters C and γ; use grid search (GridSearch) and cross-validation (Cross-Validation) to optimize hyperparameters.

[0046] Model testing: use test set to evaluate the final performance of the model; calculate performance indicators such as accuracy, recall rate, F1 score, etc. Compare the performance of different models and select the model with the best performance as the final diagnostic model.

[0047] Save the trained model as a file, load the model on new test data for prediction Output the prediction results of the model as the diagnostic results.

[0048] The specific embodiments of the application disclosed above are only used to help explain the application. The specific embodiments do not describe all the details, nor limit the application to the specific embodiments described. According to the content of the specification, many modifications and changes can be made. This specification selects and describes these embodiments in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application.

Claims

1. An obsessive-compulsive disorder diagnosis system based on multimodal brain indicators, characterized by: The system comprises: The EEG signal acquisition unit is used to acquire the EEG signal of the test subject under the preset stimulation paradigm through a multi-channel EEG acquisition device and perform preprocessing operations including bandpass filtering, independent component analysis to remove artifacts, and event-related potential segmentation extraction; A brain imaging data acquisition unit, used to synchronously acquire the tester's brain structural imaging data sMRI and brain functional imaging data fMRI; The multimodal data fusion unit is used to extract the frequency band power spectral density features and event-related potential amplitude or latency features from the EEG signal; perform Z-score normalization on the EEG features, sMRI structural features, and fMRI functional features; use a weighted average fusion algorithm to integrate the three modal features; and use a recursive feature elimination method to select the fused feature subset; The diagnostic model unit is used to input the fused feature vector into the trained support vector machine classification model and output the obsessive-compulsive disorder diagnosis result.

2. The obsessive-compulsive disorder diagnosis system based on multimodal brain indicators according to claim 1, characterized in that: The preset stimulation paradigm in the EEG signal acquisition unit includes an emotion induction paradigm or a cognitive conflict task paradigm, wherein the time window for segmented extraction of event-related potentials is set to 200-600ms after stimulation, and the P300 or N200 component is locked.

3. The obsessive-compulsive disorder diagnosis system based on multimodal brain indicators according to claim 1, characterized in that: In the brain imaging data acquisition unit, brain structural imaging data are used to extract brain area volume and cortical thickness characteristics through gray matter or white matter segmentation algorithms; brain functional imaging data are used to extract functional connection strength and local consistency indicators in resting state or task state through functional connection analysis.

4. The obsessive-compulsive disorder diagnosis system based on multimodal brain indicators according to claim 3, characterized in that: The functional connectivity analysis specifically includes: calculating the functional connectivity strength between the anterior cingulate gyrus and the basal ganglia based on the seed point analysis method, and quantifying the low-frequency amplitude fluctuations of the frontal lobe-limbic system using a local consistency algorithm.

5. The obsessive-compulsive disorder diagnosis system based on multimodal brain indicators according to claim 1, characterized in that: The frequency bands in the multimodal data fusion unit include: a delta frequency band of 1-4 Hz, a theta frequency band of 4-8 Hz, an alpha frequency band of 8-13 Hz, and a beta frequency band of 13-30 Hz.

6. The obsessive-compulsive disorder diagnosis system based on multimodal brain indicators according to claim 1, characterized in that: The multimodal data fusion unit adopts a weighted average fusion algorithm to integrate the trimodal features, including: C final =α·C EEG +β·C sMRI +γ·C fMRI in, C EEG is the output of the classifier based on EEG features; C sMRI is the output of the classifier based on MRI features; C fMRI is the output of the classifier based on fMRI features; α 、 β 、 γ is the weight coefficient, and α + β + γ =1.

7. The obsessive-compulsive disorder diagnosis system based on multimodal brain indicators according to claim 1, characterized in that: The recursive feature elimination method of the multimodal data fusion unit adopts a linear SVM as a basic model and iteratively removes the features with the smallest weights until 5-10 core feature subsets are retained.

8. The obsessive-compulsive disorder diagnosis system based on multimodal brain indicators according to claim 1, characterized in that: The data partitioning strategy of the diagnostic model unit is as follows: 70% of the training set is used for model training, 15% of the validation set is used for weight coefficient optimization and hyperparameter tuning, and 15% of the test set is used for final performance verification. Performance indicators include F1 score and AUC value.

Images