Electroencephalogram network addiction detection method based on multi-level interaction graph convolution network and related equipment

Abstract

The embodiment of the application provides a kind of based on multi-level interactive graph convolution network's brain electrical network addiction detection method and related equipment, belong to brain electrical signal processing technical field.The method comprises: obtaining the multi-channel brain electrical signal of subject, pre-processing and feature extraction are obtained node characteristics;Static brain region map and dynamic brain region map are constructed, wherein dynamic brain region map is generated by adaptive region division module, which dynamically models global functional connection between channels using multi-head attention mechanism, and learns flexible assignment matrix from node to region using graph convolution network;The static brain region map and the dynamic brain region map are input into the multi-level graph calculation unit, and the local topological features in brain region and the global topological features between brain regions are extracted respectively by parallel intra-regional graph convolution branch and inter-regional graph convolution branch;Cross-graph transition matrix is constructed by bidirectional interaction layer, to realize the bidirectional information fusion of local features and global features;The features after fusion are input into classifier, and the network addiction detection result is output.

Description

Technical Field

[0001] This application relates to the fields of electroencephalogram (EEG) signal processing and artificial intelligence-assisted diagnosis, and in particular to a method and related equipment for detecting EEG addiction based on a multi-level interactive graph convolutional network. Background Technology

[0002] Internet addiction (IA) has become a growing public health problem worldwide, with a particularly high incidence among adolescents. Studies have shown that IA is closely related to various psychological and behavioral abnormalities, including impulse control disorders, attention deficit, and decreased self-esteem, severely impacting their quality of life, work, and studies. Currently, the clinical diagnosis of IA primarily relies on questionnaires or professional interviews, such as the Internet Addiction Test (IAT). However, these methods are susceptible to influences such as individual subjectivity, self-concealment, and test sensitivity, making it difficult to guarantee the objectivity and accuracy of the diagnosis.

[0003] Electroencephalography (EEG), as a non-invasive, high-temporal-resolution, and low-cost neurophysiological monitoring method, can capture changes in the brain's electrical activity in real time, and its signals are not easily affected by subjective consciousness. Therefore, it is widely regarded as an important tool for assisted detection of intracranial pressure (IA). In recent years, more and more studies have attempted to combine EEG signals with deep learning techniques to achieve automatic identification of IA. Traditional methods mainly rely on manual feature engineering and shallow classifiers (such as Support Vector Machines (SVM) and Random Forests (RF). However, these methods are highly dependent on expert knowledge during feature extraction and struggle to uncover deep nonlinear patterns in the data.

[0004] With the rapid development of deep learning, Graph Neural Networks (GNNs) have been increasingly introduced into EEG modeling due to their advantages in processing non-Euclidean structured data. However, existing GNN-based EEG analysis methods still face the following technical bottlenecks: 1) Static brain region segmentation cannot adapt to dynamic functional connectivity patterns. Existing methods typically divide electrodes into several fixed brain regions (such as the frontal lobe, parietal lobe, occipital lobe, etc.) based on prior neuroscience knowledge to model local and global brain functional connectivity. However, the brain networks of IA patients exhibit highly dynamic functional reorganization characteristics, and fixed segmentation methods cannot capture inter-individual differences and real-time changes between brain regions, limiting the model's expressive power and generalization performance.

[0005] 2) Neglecting the interaction mechanisms between brain regions. The brain's functions depend on the coordination and information exchange between different brain regions. Existing research mostly focuses on modeling brain regions at a single level (such as within or between regions), lacking modeling of the interaction mechanisms between information at different levels, making it difficult to fully characterize the complex abnormal patterns of brain functional connectivity in IA patients.

[0006] 3) Lack of effective modeling of individual differences. Because the brain functional reorganization patterns of IA patients vary from person to person, traditional partitioning methods based on fixed templates are difficult to reflect the individual-specific brain region combinations, resulting in decreased recognition performance of the model when faced with new individuals, making it difficult to meet the requirements of high generalization ability for clinical applications. Summary of the Invention

[0007] The main objective of this application is to propose a brain-based electroencephalogram (EEG) addiction detection method, system, electronic device, storage medium, and program product based on a multi-level interactive graph convolutional network. This aims to solve problems such as static brain region segmentation, poor adaptability to individual differences, and lack of brain region interaction mechanisms in the prior art, thereby significantly improving the accuracy and interpretability of EEG-based IA recognition tasks.

[0008] To achieve the above objectives, one aspect of this application proposes a brainwave addiction detection method based on a multi-level interactive graph convolutional network, the method comprising: The subject's multi-channel EEG signals were acquired, and the multi-channel EEG signals were preprocessed and feature extracted to obtain the node features of each channel. Based on the node features, static brain region maps and dynamic brain region maps are constructed. The dynamic brain region map is generated by an adaptive region partitioning module. The adaptive region partitioning module uses a multi-head attention mechanism to dynamically model the global functional connections between channels and uses a graph convolutional network to learn a flexible allocation matrix from nodes to regions in order to construct individualized dynamic brain regions. The static brain region map and the dynamic brain region map are respectively input into a multi-level graph computing unit. The multi-level graph computing unit includes parallel intra-region graph convolution branches and inter-region graph convolution branches, which are used to extract local topological features within brain regions and global topological features between brain regions, respectively. Cross-graph information fusion is performed on the local topological features and the global topological features through a bidirectional interaction layer. The bidirectional interaction layer constructs a cross-graph transition matrix to realize bidirectional information transmission and updating between the local topological features and the global topological features. The fused features are input into a classifier, which outputs the internet addiction detection results for the subject.

[0009] In some embodiments, the adaptive region partitioning module specifically includes: A multi-head attention mechanism is applied to the node features to generate a relation matrix, and a global connectivity matrix is ​​calculated by combining the learnable weight vectors. By using a graph convolutional network and the Softmax function, a probability assignment matrix is ​​learned, the elements of which represent the probability of assigning a channel node to a dynamic region; Based on the probability allocation matrix, the adjacency matrix and feature matrix at the dynamic region level are reconstructed; An allocation regularization term is introduced, which includes a Frobenius norm regularization term and an entropy-based sparsity constraint, to ensure the consistency between functional connectivity strength and channel region allocation probability, and to reduce allocation ambiguity.

[0010] In some embodiments, in the multi-level graph calculation unit: The intra-region graph convolution branch treats each divided local brain region as an independent graph and performs graph convolution operation, and concatenates and splices the convolution results of all local brain regions to obtain a complete intra-region feature representation; The interval graph convolution branch performs graph convolution operations on the adjacency matrix and feature matrix at the dynamic region level, and outputs a macroscopic interval feature representation.

[0011] In some embodiments, the bidirectional interaction layer specifically includes: Project the local topological features and the global topological features into a compatible latent space; A cross-graph transition matrix is ​​generated by calculating the dot product and normalizing along the rows using Softmax. The elements of the cross-graph transition matrix represent the weights of the normalization effect of the region on the node. The cross-graph transition matrix is ​​used to adaptively fuse macroscopic interval features into microscopic intra-region features, and the intra-region features are fused back into interval features through transpose mapping, thereby achieving bidirectional information diffusion.

[0012] In some embodiments, the preprocessing and feature extraction include: The multi-channel EEG signals were subjected to bandpass filtering and independent component analysis for noise reduction to obtain clean EEG signals. The pure EEG signal is segmented into short time segments, and the power spectral density of each segment in multiple frequency bands of each channel is extracted as node features.

[0013] In some embodiments, the classifier is a multilayer perceptron; the internet addiction detection result includes the classification probability of normal or addictive.

[0014] To achieve the above objectives, another aspect of this application proposes a brainwave addiction detection system based on a multi-level interactive graph convolutional network, the system comprising: The data acquisition module is used to acquire multi-channel EEG signals from the subjects; The preprocessing and feature extraction module is used to preprocess and extract features from the multi-channel EEG signals to obtain the node features of each channel. An adaptive region partitioning module is used to construct a dynamic brain region map based on the node features. The adaptive region partitioning module adopts a multi-head attention mechanism to dynamically model the global functional connections between channels and uses a graph convolutional network to learn a flexible allocation matrix from nodes to regions in order to construct individualized dynamic brain regions. The static brain region construction module is used to construct static brain region maps based on prior knowledge. The multi-level graph computation unit includes parallel intra-region graph convolution branches and inter-region graph convolution branches, which are used to extract local topological features within brain regions and global topological features between brain regions, respectively. A bidirectional interaction layer is used to perform cross-graph information fusion between the local topological features and the global topological features. The bidirectional interaction layer constructs a cross-graph transition matrix to realize bidirectional information transmission and updating between the local topological features and the global topological features. The classification module is used to input the fused features into the classifier and output the internet addiction detection results of the subject.

[0015] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0016] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0017] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0018] The embodiments of this application include at least the following beneficial effects: This application provides a brain-computer interface (BCI) addiction detection method, system, electronic device, storage medium, and program product based on a multi-level interactive graph convolutional network. First, an adaptive region partitioning module is used to automatically learn the dynamic graph structure in a data-driven manner. Second, a multi-graph computation unit is used to combine static and dynamic graphs to capture multi-level spatial information. Finally, a bidirectional interaction layer is introduced to achieve bidirectional communication and updating of features within and between brain regions. Through the above mechanisms, this application significantly improves the accuracy and robustness of BCI identification. Attached Figure Description

[0019] Figure 1This is a flowchart of the brainwave addiction detection method based on a multi-level interactive graph convolutional network provided in the embodiments of this application; Figure 2 This is a schematic diagram of the product application scenario and cloud-edge collaborative workflow of this application embodiment; Figure 3 This is a schematic diagram of the overall framework of the model in an embodiment of this application; Figure 4 This is a schematic diagram of the bidirectional interaction layer structure according to an embodiment of this application; Figure 5 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0022] Most existing internet addiction identification technologies rely on simple machine learning or deep learning algorithms to extract features. This first existing technology combines electroencephalography (EEG) with internet addiction detection, aiming to address the problem of traditional internet addiction diagnosis relying on subjective questionnaires and lacking objective physiological indicators. Its core technical solution mainly includes the following steps: First, EEG signals from test subjects with different levels of internet addiction are collected, and the raw EEG signals are filtered and preprocessed to remove noise and artifacts. Second, the energy features of the filtered EEG signals in different frequency bands are calculated to construct an energy feature matrix, which serves as the input to the subsequent classification model. Then, an Extreme Learning Machine (ELM) based on a deep learning mapping kernel function is used to train the energy feature matrix, obtaining a classification model for judging the degree of internet addiction. Finally, the test subject's EEG signal, after the same preprocessing and feature extraction, is input into the trained classification model, outputting the result of their internet addiction level assessment. This solution, by introducing a kernel-optimized ELM model, achieves faster training speed and stronger generalization ability compared to traditional neural networks while maintaining learning accuracy, providing physicians with an objective auxiliary diagnostic solution for judging internet addiction.

[0023] While the aforementioned existing technologies have made some progress in the objective auxiliary diagnosis of internet addiction, significant technical limitations remain when dealing with complex brain network mechanisms and the widespread individual differences among subjects in practical applications. A detailed analysis follows: 1) Ignoring individual differences among subjects and the dynamic reorganization characteristics of brain networks leads to poor generalization ability. Existing technologies typically rely on fixed brain region divisions or independent channel feature extraction based on prior knowledge. Specifically, the physiological characteristics of different individuals vary greatly, and inter-individual neural networks (IAs) cause highly dynamic reorganization of brain functional networks. This static partitioning model cannot adapt to the dynamic changes and individual variability in brain functional connections among different individuals. When the model is applied to new subjects with different characteristics, the model's generalization ability is limited, and accuracy tends to decline. Our solution: This application introduces an adaptive region partitioning module. This module is not limited to static partitioning but utilizes a global multi-head attention mechanism to automatically learn the dynamic combination of electrode channels in a data-driven manner, constructing exclusive dynamic brain regions for different individuals. This effectively mitigates the impact of individual differences and can accurately capture dynamic topological dependencies.

[0024] 2) Isolated extraction of local features, lacking joint modeling of brain spatial topological dependencies. Existing technologies mainly extract features from individual channels (such as frequency band energy) and directly splice them, essentially treating EEG signals as isolated time series. Specifically, intracranial neural network (IA) not only affects local brain region activity but also leads to abnormal functional connectivity between different brain regions. Existing technologies sever these spatial connections, losing the crucial high-dimensional biomarker feature of "brain functional connectivity," thus limiting the upper limit of disease detection. Our solution: This application proposes a multi-level graph computation unit. Combining static and dynamic brain region segmentation, it simultaneously performs deep modeling of topological connectivity within and between brain regions through parallel graph convolutional network branches. This multi-level graph modeling can comprehensively capture the structural abnormalities in the brain's spatial network caused by IA.

[0025] 3) The feature fusion method is singular, severing the dynamic interaction mechanism between local and global brain regions. Existing methods typically input features directly into the classifier after extraction, lacking a characterization of the bidirectional flow of information between different levels of brain regions. Specifically, the human brain is essentially a dynamic collaborative system that continuously interacts through bidirectional information flow. Activation of a specific brain region is often accompanied by the co-activation of other regions. The lack of cross-level (microscopic local and macroscopic global) dynamic feature interaction makes it difficult for models to effectively learn the complex bidirectional relationships between brain regions. Solution: This application innovatively designs a bidirectional interaction layer. This layer constructs a cross-graph transition matrix in the multi-graph computation unit, enabling adaptive bidirectional information transfer and correction between macroscopic region features and microscopic intra-region features. This realistically simulates the neurobiological principle of continuous information exchange in the brain, achieving adaptive cross-graph information fusion.

[0026] In summary, existing technologies have not yet formed a unified framework that can effectively integrate prior knowledge and dynamic learning mechanisms, take into account both local and global information, and model the interaction relationships between brain regions. In view of this, this application provides a brain-electrical addiction detection method, system, electronic device, storage medium, and program product based on a multi-level interactive graph convolutional network. This scheme constructs a learning framework that integrates static and dynamic brain region segmentation, aiming to simultaneously solve the problems of static brain region segmentation, significant individual differences, and lack of brain region interaction mechanisms in IA detection, thereby significantly improving the accuracy and interpretability of the model in EEG-based IA recognition tasks. The main technical means of this application include the following aspects: 1) Adaptive brain region segmentation module: Addressing the problem that traditional methods rely on fixed prior brain region segmentation and cannot adapt to the dynamic reorganization of brain function in IA patients, this application proposes a data-driven adaptive brain region segmentation mechanism. This module dynamically models the global functional connectivity between EEG channels through a multi-head attention mechanism and utilizes a graph neural network to learn a flexible allocation matrix from nodes to regions, achieving automatic construction of individualized brain regions. By introducing regularization terms based on functional connectivity strength and entropy constraints, the partitioning results are ensured to be both biologically interpretable and flexibly adaptable to the heterogeneity of brain networks among different individuals, thereby effectively alleviating the problem of distribution differences across subjects. 2) Multi-level graph computation unit: To comprehensively characterize the multi-level abnormal patterns of brain functional connectivity in IA patients, this application designs a parallel bi-branch graph convolutional structure to model functional connectivity within and between brain regions. In the intra-region branch, functional connectivity graphs are independently constructed for each brain region and local features are extracted; in the inter-region branch, a global graph is constructed based on region-level features to model high-order interaction relationships between brain regions. Through this hierarchical modeling strategy, the model can simultaneously capture local functional specificity and global information integration capabilities, improving the representation ability of IA-related brain activity patterns. 3) Bidirectional interaction layer: To address the problem that existing methods neglect information flow between brain regions, this application proposes a cross-graph information fusion mechanism. By constructing a learnable node-region transition matrix, bidirectional interaction and information diffusion between intra-region features and inter-region features are realized. This mechanism allows local details to be passed up to the region-level representation, while global contextual information guides local feature learning downwards, forming a complementary and enhanced collaborative optimization process. Through multi-layered iterative interactions, the model can more comprehensively simulate the brain's dynamic collaborative mechanisms, further improving the accuracy and robustness of IA recognition.

[0027] The brain-electroelectric addiction detection method provided in this application relates to the field of EEG signal processing and artificial intelligence-assisted diagnosis technology. The brain-electroelectric addiction detection method provided in this application can be applied to a terminal, a server, or software running on a terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or in-vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the brain-electroelectric addiction detection method, but is not limited to the above forms.

[0028] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0029] like Figure 1 As shown, this embodiment provides a brainwave addiction detection method based on a multi-level interactive graph convolutional network, specifically including the following steps: Step S1: Acquire multi-channel EEG signals from the subject, and preprocess and extract features from the multi-channel EEG signals to obtain the node features of each channel; Step S2: Based on node features, construct static brain region maps and dynamic brain region maps. The dynamic brain region map is generated by an adaptive region partitioning module. The adaptive region partitioning module uses a multi-head attention mechanism to dynamically model the global functional connections between channels and uses a graph convolutional network to learn a flexible allocation matrix from nodes to regions in order to construct individualized dynamic brain regions. Step S3: Input the static brain region map and the dynamic brain region map into the multi-level graph computing unit respectively. The multi-level graph computing unit includes parallel intra-region graph convolution branches and inter-region graph convolution branches, which are used to extract local topological features within brain regions and global topological features between brain regions, respectively. Step S4: Cross-graph information fusion of local and global topological features is performed through a bidirectional interaction layer. The bidirectional interaction layer constructs a cross-graph transition matrix to realize bidirectional information transfer and updating between local and global topological features. Step S5: Input the fused features into the classifier and output the test results of the subject's internet addiction.

[0030] The solutions of the embodiments of this application will be described in detail below with reference to the accompanying drawings and specific application examples.

[0031] (1) Product side 1.1) Application Scenarios and Product Forms The technical solution proposed in this embodiment is primarily aimed at application scenarios such as auxiliary diagnosis in clinical mental health, psychological screening of adolescents in schools, and early prevention and intervention of internet addiction in families. Its core product form is a "portable multi-channel EEG acquisition device," supplemented by a matching smart app for doctors / users and a cloud-based EEG analysis platform. Traditional internet addiction screening heavily relies on self-report questionnaires completed subjectively by the subjects or interviews with professional physicians, which is not only time-consuming but also highly susceptible to patient concealment and cognitive biases, resulting in low sensitivity. In contrast, this solution allows subjects to complete objective physiological signal monitoring in a comfortable and quiet state, such as in a clinic or psychological counseling room. Utilizing high-performance computing power in the cloud to run algorithms, it can keenly capture and adapt to individual differences in the reorganization of brain functional networks among different subjects, thereby providing highly accurate early warnings of internet addiction risk.

[0032] 1.2) Cloud-device collaborative workflow Considering that the core algorithm of this embodiment involves the construction of a complex multi-level brain network topology, bidirectional information interaction, and high-dimensional matrix operations, it places high demands on computing resources. Therefore, this solution designs a cloud-edge collaborative data processing flow of "lightweight data acquisition on the edge + heavy computing in the cloud," see [link to relevant documentation]. Figure 2 The specific steps are as follows: Step 1: End-to-End Data Acquisition. The subject wears a multi-channel EEG acquisition device in a clinic or quiet environment. The device continuously acquires EEG signals for approximately 5 to 8 minutes while the subject is at rest with their eyes closed. The device's built-in microcontroller performs analog-to-digital conversion and preliminary analog amplification of the raw signals. The raw EEG data is then synchronized in real-time or offline to the accompanying doctor's / user's mobile terminal via a secure, encrypted wireless network (such as Wi-Fi or Bluetooth Low Energy).

[0033] Step Two: Cloud-based Brain Network Map Construction and Intelligent Reasoning. The mobile terminal uploads the desensitized resting-state EEG data of the subjects to the "Internet Addiction Diagnosis Cloud Platform".

[0034] Data preprocessing and feature extraction: The cloud server first performs bandpass filtering on the raw data from 1 to 45 Hz and uses independent component analysis (ICA) to remove non-EEG artifacts (such as EEG and EMG interference). Then, the continuous data is segmented into short time segments (e.g., 2 seconds), and the power spectral density (PSD) of multiple frequency bands is extracted as node features.

[0035] Dual-track graph construction and multi-level inference: The core algorithm is pre-trained and deployed in the cloud. When new data is received, the system first divides the channels into standard static brain regions (such as the frontal lobe and parietal lobe) based on prior knowledge. At the same time, the adaptive region segmentation module automatically learns and generates a unique dynamic brain region allocation matrix for the current subject through a global attention mechanism, so as to dynamically adapt to and eliminate individual-specific differences in brain network reorganization. Subsequently, the multi-graph computing unit, combined with the bidirectional interaction layer, extracts the topological features within the micro-brain regions and between the macro-brain regions in parallel, and performs bidirectional information fusion and diffusion across graphs.

[0036] Accurate classification: After cascading multi-level graph features, the fully connected classifier outputs a prediction of the subject's probability of network addiction (normal / addictive).

[0037] Step 3: Results Feedback and Interaction (User Interface): The cloud transmits the final analysis results back to the terminal APP. The APP interface not only intuitively displays the subject's "Comprehensive Risk Report of Internet Addiction," but also generates a clinically interpretable "Brain Functional Connectivity Atlas" based on the topological weights calculated by the algorithm. The system highlights key abnormal electrode pairs and abnormal connectivity pathways (such as abnormal connections between the frontal lobe, parietal lobe, and occipital lobe) in the atlas, thereby providing intuitive and objective neurobiological evidence for further diagnosis and psychological intervention by physicians.

[0038] 1.3) User operation and interaction methods Wearing: Subjects wore the multi-channel portable EEG cap on their heads in a quiet environment, ensuring that all electrode contacts (especially electrodes covering key areas such as the frontal lobe, parietal lobe, temporal lobe, and occipital lobe) were in full contact with the scalp to obtain high-quality EEG signals.

[0039] Start-up: Subjects remain in a relaxed, resting state with their eyes closed. Doctors or users activate the device by clicking the "Start Resting State Acquisition" button on the smart app. The entire EEG signal acquisition process only takes 5 to 8 minutes.

[0040] View the report: After the data collection is completed, the EEG data is automatically synchronized to the cloud for processing. The APP homepage will intuitively display the subject's predicted IA probability. Doctors or users can further click to view a detailed brain functional connectivity map generated by the algorithm of this invention. This map can accurately reflect the dynamic abnormal functional connectivity status of the subject's local and global brain (such as key connections between the frontal lobe, parietal lobe, and occipital lobe), providing visualized neurophysiological evidence for the auxiliary diagnosis of internet addiction.

[0041] (2) Technical side The following is the theoretical framework and main components of the multi-level interactive graph convolutional network based on brain region segmentation proposed in this embodiment for internet addiction detection. Figure 3 As shown, the overall processing logic of this solution is as follows: 1. Input phase: Acquire multichannel electroencephalogram (EEG) data of subjects in a resting state.

[0042] 2. Preprocessing and Feature Extraction Stage: The raw EEG signal is denoised by bandpass filtering and independent component analysis (ICA). Subsequently, the continuous signal is segmented into standard short time segments, and the power spectral density (PSD) of multiple neurophysiologically relevant frequency bands is extracted to construct initial channel node features.

[0043] 3. Dual-track brain map construction stage: On the one hand, static brain region division is performed based on prior neuroscience knowledge to generate local feature maps of predefined regions; on the other hand, an adaptive region division module is used to learn the electrode allocation probability in a data-driven manner through a global attention mechanism, and automatically reconstruct a dynamic region map that adapts to individual differences.

[0044] 4. Multi-level Interactive Graph Computation Stage: Static and dynamic mind maps are input into parallel multi-graph computing units, and high-order topological dependencies within regions (local) and between regions (global) are extracted using graph convolution branches. During feature extraction, a cross-graph transition matrix is ​​innovatively calculated through a bidirectional interactive layer, achieving adaptive bidirectional information fusion between features within micro-regions and features between macro-regions.

[0045] 5. Model Training and Optimization Phase: The static and dynamic features updated through multi-level interactions are concatenated and input into a multilayer perceptron to output predicted probabilities. To ensure the interpretability and stability of dynamic partitioning, cross-entropy classification loss is used. Based on this, a feature assignment consistency regularization term was introduced. With entropy sparsity constraints Perform joint optimization. The overall loss function is expressed as:

[0046] 6. Inference stage: Input the EEG data of the target subject to be tested into the trained end-to-end graphical neural network, and output the final classification prediction result for IA (e.g., normal / addictive).

[0047] 2.1) Data Preprocessing Data preprocessing is a fundamental step in eliminating physiological noise and standardizing the distribution of EEG data, providing high-quality node features for subsequent graph networks. The processing flow for multi-channel EEG signals in this solution is as follows: Signal filtering and denoising: After acquiring the raw resting-state multichannel EEG data of the subjects, a Butterworth bandpass filter of 1-45 Hz was first applied to retain the core frequency bands closely related to neurophysiological activity. Subsequently, independent component analysis (ICA) algorithm was used to further remove artifacts and residual noise such as electrooculography (EOG) and electromyography (EMG).

[0048] Data Segmentation and Feature Construction: Continuous, clean EEG data is segmented into non-overlapping 2-second segments, which are then used as independent input samples for augmentation. For each channel node of each segment, the power spectral density (PSD) of five classic frequency bands is extracted as node features, specifically including: (0.5-4Hz) (4-8Hz) (8-13Hz) (13-30Hz) and (30-50Hz). This ultimately forms the initial feature matrix of the nodes. ,in For the number of channels, For feature dimensions.

[0049] 2.2) Adaptive Region Partitioning Module To overcome the limitations of fixed manual brain region segmentation in adapting to dynamic network reorganization and individual variations among different patients, this scheme designs an adaptive region segmentation module to dynamically construct region-level graph structures in a data-driven manner. Global relation extraction: First, the feature matrix... A multi-head attention mechanism with six heads is applied to generate a relation matrix. Combined with learnable weight vectors The global connectivity matrix reflecting the dynamic topology between channels is calculated: Dynamic allocation matrix learning: Utilizing a single-layer graph convolutional network (GCN) and the softmax function, a probability allocation matrix is ​​learned. The formula is ,in and For learnable parameters, This represents the preset number of dynamic regions. Matrix elements. Indicates that the node Allocate to dynamic region The probability of reconstructing the region-level graph: based on the allocation matrix. Reconstruct the adjacency matrix at the macro-dynamic regional level. and characteristic matrix The calculation methods are as follows: , Assignment regularization: To ensure stable training and conformity to biological principles, two regularization constraints are introduced. The first is the Frobenius norm regularization term. This ensures that the functional connectivity strength is highly consistent with the regional allocation probability of the channels. The second term is an entropy-based sparsity constraint. This ensures that each channel is clearly assigned to a specific main functional area, reducing ambiguity in allocation.

[0050] 2.3) Multi-level graph calculation unit The multi-level graph computation unit extracts spatial topological features at both the micro (inter-channel) and macro (inter-brain) scales through a two-branch graph convolutional network. Intra-region graph convolution: In this branch, the algorithm treats each pre-defined local brain region as an independent graph and performs GCN convolution operations. The convolution operation is defined as follows: ,in For Laplace matrix, The activation function is non-linear. The extraction results from all local brain regions are concatenated and concatenated to obtain a complete representation of the intra-region features. Interval Graph Convolution: Another parallel branch performs GCN convolution on the previously aggregated region-level graphs to capture high-order dynamic dependencies and functional integration between different large-scale brain regions, outputting macroscopic interval features. .

[0051] 2.4) Bidirectional Interaction Layer like Figure 4 As shown, to simulate the information exchange mechanism of the human brain, this scheme inserts a bidirectional interaction layer between the GCN levels of the multi-level graph computation unit to establish bidirectional influence between intra-region and inter-region features. Cross-graph transition matrix calculation: First, intra-region features... and interval features Projecting to a compatible potential space: , Next, cross-graph transition moments are generated by calculating the dot product and performing softmax normalization along the rows. The formula is: Matrix elements Characterizes the region For nodes Normalization affects weights. Bidirectional feature update and diffusion: Utilizing... The macroscopic interval features are adaptively propagated and fused into the microscopic intra-region features: ,in These are learnable weights. Similarly, information from intra-region features is also fused back into the inter-region features through a corresponding transpose mapping. This bidirectional fusion effectively prevents the isolation of mind maps at different levels and improves the sensitivity of identifying abnormal connections in network addiction.

[0052] 2.5) Model Optimization Objective and Classification Inference After completing multi-level interactive learning of both dynamic and static maps, the system will output features from the static branch. With dynamic branch output characteristics Perform cascading splicing: Subsequently, The data is fed into a classifier consisting of two multilayer perceptrons (MLPs), which outputs the network's predicted addiction probability distribution for each sample. The final loss function during network training is... Classification loss based on cross-entropy Together with the allocation regularization terms of the two aforementioned adaptive region partitioning modules, it constitutes the following form:

[0053] (3) Advantages and beneficial effects In summary, this application significantly overcomes the technical bottlenecks of existing technologies in static brain region segmentation, individual difference adaptation, and cross-level brain function interaction by constructing a dual-track learning framework that integrates prior neuroscience knowledge and data-driven approaches. Specific beneficial effects are as follows: 1) Significantly improved model generalization ability and individual adaptability in cross-subject scenarios. This application innovatively introduces an adaptive region segmentation module, which automatically learns the dynamic combination of electrode channels for specific subjects in a data-driven manner, rather than relying on fixed manual brain region segmentation. This personalized dynamic graph construction effectively mitigates the feature distribution shift caused by physiological differences and brain functional network reorganization among different subjects, enabling the model to adapt to dynamic topological changes when faced with EEG data from new, unseen subjects, greatly improving the model's generalization ability and robustness in real-world cross-subject application scenarios.

[0054] 2) This approach comprehensively captures multi-level topological dependencies in the brain, achieving more accurate feature representation. Addressing the shortcomings of existing methods that often isolate local features and sever the interaction between local and global brain regions, this application designs a multi-level graph computation unit and a bidirectional interaction layer. By simultaneously extracting microscopic (intra-region) and macroscopic (inter-region) features through parallel graph convolutional networks and utilizing cross-graph transition matrices to achieve adaptive bidirectional information fusion of features, this technology realistically simulates the continuous cross-regional information exchange and collaboration mechanism of the human brain. This multi-level interaction architecture significantly improves the model's ability to capture abnormal brain function features in internet addiction patients.

[0055] 3) Achieved industry-leading detection accuracy on real clinical datasets. To verify the effectiveness of this approach, extensive comparative experiments were conducted on the real clinical EEG dataset (DNAP) and the public dataset (LEMON). Experimental results show that the proposed method comprehensively outperforms existing shallow machine learning models and advanced graph neural network models. Specifically, this model achieved high accuracies of 97.46% and 96.88% on the DNAP and LEMON datasets, respectively, and F1 scores of 96.60% and 95.84%. Compared to the existing best-in-class Pyramid Graph Convolutional Network (PGCN), this approach significantly improves accuracy by 3.60% on the DNAP dataset, effectively overcoming the common class imbalance problem in clinical data and meeting the high standards of medical-grade assisted diagnosis.

[0056] 4) It provides extremely high clinical biological interpretability and visual diagnostic value. Traditional deep learning models are often considered "black boxes," while this application demonstrates excellent interpretability. By extracting the assignment matrix and attention weights of the multi-level graph topology within the model, the system can directly generate a "brain functional connectivity topology map" of the subject, accurately locating core brain regions (such as the frontal and parietal lobes) highly associated with internet addiction, as well as key abnormal electrode pair interactions. This not only provides objective neuroimaging evidence for verifying the pathological mechanism of IA, but also provides clinical psychiatrists with intuitive and visual biomarkers, changing the previous diagnostic dilemma of relying solely on subjective questionnaires.

[0057] This application also provides a brainwave addiction detection system based on a multi-level interactive graph convolutional network, the system comprising: The data acquisition module is used to acquire multi-channel EEG signals from the subjects; The preprocessing and feature extraction module is used to preprocess and extract features from the multi-channel EEG signals to obtain the node features of each channel. An adaptive region partitioning module is used to construct a dynamic brain region map based on the node features. The adaptive region partitioning module adopts a multi-head attention mechanism to dynamically model the global functional connections between channels and uses a graph convolutional network to learn a flexible allocation matrix from nodes to regions in order to construct individualized dynamic brain regions. The static brain region construction module is used to construct static brain region maps based on prior knowledge. The multi-level graph computation unit includes parallel intra-region graph convolution branches and inter-region graph convolution branches, which are used to extract local topological features within brain regions and global topological features between brain regions, respectively. A bidirectional interaction layer is used to perform cross-graph information fusion between the local topological features and the global topological features. The bidirectional interaction layer constructs a cross-graph transition matrix to realize bidirectional information transmission and updating between the local topological features and the global topological features. The classification module is used to input the fused features into the classifier and output the internet addiction detection results of the subject.

[0058] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0059] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0060] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0061] Please see Figure 5 , Figure 5 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 501 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 502 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 502 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 502 and is called and executed by the processor 501 using the methods described in the embodiments of this application. The input / output interface 503 is used to implement information input and output; The communication interface 504 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 505 transmits information between various components of the device (e.g., processor 501, memory 502, input / output interface 503, and communication interface 504); The processor 501, memory 502, input / output interface 503, and communication interface 504 are connected to each other within the device via bus 505.

[0062] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0063] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0064] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0065] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0066] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented in the embodiments of this program product are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. The executable computer program code or "code" used to perform the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0067] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0068] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0069] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0070] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0071] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0072] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0073] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0074] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0075] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0076] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0077] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A brainwave addiction detection method based on a multi-level interactive graph convolutional network, characterized in that, The method includes the following steps: The subject's multi-channel EEG signals were acquired, and the multi-channel EEG signals were preprocessed and feature extracted to obtain the node features of each channel. Based on the node features, static brain region maps and dynamic brain region maps are constructed. The dynamic brain region map is generated by an adaptive region partitioning module. The adaptive region partitioning module uses a multi-head attention mechanism to dynamically model the global functional connections between channels and uses a graph convolutional network to learn a flexible allocation matrix from nodes to regions in order to construct individualized dynamic brain regions. The static brain region map and the dynamic brain region map are respectively input into a multi-level graph computing unit. The multi-level graph computing unit includes parallel intra-region graph convolution branches and inter-region graph convolution branches, which are used to extract local topological features within brain regions and global topological features between brain regions, respectively. Cross-graph information fusion is performed on the local topological features and the global topological features through a bidirectional interaction layer. The bidirectional interaction layer constructs a cross-graph transition matrix to realize bidirectional information transmission and updating between the local topological features and the global topological features. The fused features are input into a classifier, which outputs the internet addiction detection results for the subject.

2. The method according to claim 1, characterized in that, The adaptive region partitioning module specifically includes: A multi-head attention mechanism is applied to the node features to generate a relation matrix, and a global connectivity matrix is ​​calculated by combining the learnable weight vectors. By using a graph convolutional network and the Softmax function, a probability assignment matrix is ​​learned, the elements of which represent the probability of assigning a channel node to a dynamic region; Based on the probability allocation matrix, the adjacency matrix and feature matrix at the dynamic region level are reconstructed; An allocation regularization term is introduced, which includes a Frobenius norm regularization term and an entropy-based sparsity constraint, to ensure the consistency between functional connectivity strength and channel region allocation probability, and to reduce allocation ambiguity.

3. The method according to claim 1, characterized in that, In the multi-level graph calculation unit: The intra-region graph convolution branch treats each divided local brain region as an independent graph and performs graph convolution operation, and concatenates and splices the convolution results of all local brain regions to obtain a complete intra-region feature representation; The interval graph convolution branch performs graph convolution operations on the adjacency matrix and feature matrix at the dynamic region level, and outputs a macroscopic interval feature representation.

4. The method according to claim 1, characterized in that, The bidirectional interaction layer specifically includes: Project the local topological features and the global topological features into a compatible latent space; A cross-graph transition matrix is ​​generated by calculating the dot product and normalizing along the rows using Softmax. The elements of the cross-graph transition matrix represent the weights of the normalization effect of the region on the node. The cross-graph transition matrix is ​​used to adaptively fuse macroscopic interval features into microscopic intra-region features, and the intra-region features are fused back into interval features through transpose mapping, thereby achieving bidirectional information diffusion.

5. The method according to claim 1, characterized in that, The preprocessing and feature extraction include: The multi-channel EEG signals were subjected to bandpass filtering and independent component analysis for noise reduction to obtain clean EEG signals. The pure EEG signal is segmented into short time segments, and the power spectral density of each segment in multiple frequency bands of each channel is extracted as node features.

6. The method according to claim 1, characterized in that, The classifier is a multilayer perceptron; the network addiction detection result includes the classification probability of normal or addictive.

7. A brainwave addiction detection system based on a multi-level interactive graph convolutional network, characterized in that, include: The data acquisition module is used to acquire multi-channel EEG signals from the subjects; The preprocessing and feature extraction module is used to preprocess and extract features from the multi-channel EEG signals to obtain the node features of each channel. An adaptive region partitioning module is used to construct a dynamic brain region map based on the node features. The adaptive region partitioning module adopts a multi-head attention mechanism to dynamically model the global functional connections between channels and uses a graph convolutional network to learn a flexible allocation matrix from nodes to regions in order to construct individualized dynamic brain regions. The static brain region construction module is used to construct static brain region maps based on prior knowledge. The multi-level graph computation unit includes parallel intra-region graph convolution branches and inter-region graph convolution branches, which are used to extract local topological features within brain regions and global topological features between brain regions, respectively. A bidirectional interaction layer is used to perform cross-graph information fusion between the local topological features and the global topological features. The bidirectional interaction layer constructs a cross-graph transition matrix to realize bidirectional information transmission and updating between the local topological features and the global topological features. The classification module is used to input the fused features into the classifier and output the internet addiction detection results of the subject.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images