A method and system for auxiliary diagnosis of autism based on two-person brain functional connectivity maps and graph attention networks.

Abstract

This invention discloses an auxiliary diagnostic method and system for autism based on a dual-person brain functional connectivity map and graph attention network. It comprises a dual-person fNIRS hyperscanning data acquisition module, a preprocessing and graph construction module, and an edge-weighted enhanced graph attention network classification module. The system has the following technical features: synchronously acquiring brain activity signals of the subject and examinee during social interactions using hyperscanning technology; constructing a dual-person brain functional connectivity map including nodes in both brain regions and the strength of inter-brain connections; and dynamically aggregating and learning node features and edge weights in the graph using an edge-weighted enhanced graph attention network. This invention can objectively capture abnormal inter-brain synchronization in autistic individuals during real social interactions, providing automated, rapid, and quantifiable auxiliary diagnostic results, effectively improving the objectivity and accuracy of diagnosis, and reducing reliance on professional clinical experience.

Description

Technical Field

[0001] This invention relates to the fields of medical auxiliary diagnosis and brain-computer interface technology, specifically to an autism auxiliary diagnosis method and system based on a dual-brain functional connectivity map and graph attention network. Background Technology

[0002] Currently, the diagnosis of autism spectrum disorder (ASD) is mainly based on standards such as the Diagnostic and Statistical Manual of Mental Disorders (DSM), through clinical observation, parent interviews, and structured scales (such as ADOS). This approach has the following technical limitations: 1. High subjectivity and insufficient consistency: Diagnosis relies heavily on the assessor's experience and subjective judgment, easily influenced by situational, emotional, and cultural factors, leading to insufficient diagnostic consistency, especially for mild or atypical cases. 2. Neglect of two-way interaction mechanisms: Most current technologies only analyze the patient's unilateral brain activity, failing to characterize the neural coordination mechanisms in the two-way process of social interaction. The core symptom of ASD is social communication impairment, and its essential defect lies in the abnormal neural synchronicity of the two parties in the interaction; unilateral analysis cannot effectively capture this key feature. 3. High professional threshold and uneven resource distribution: ASD diagnosis heavily relies on the long-term experience of child psychiatrists and other professionals. The long training cycle for professionals and their uneven geographical distribution result in long waiting times for diagnosis in many areas, lacking a technological tool that can provide objective and rapid auxiliary screening.

[0003] Functional near-infrared spectroscopy (fNIRS), as a non-invasive optical brain imaging technique, has good resistance to motion interference and is suitable for acquiring brain activity signals in natural interactive scenarios. However, how to use fNIRS technology to objectively and quantitatively analyze the brain functional connectivity patterns of two people in social interactions and use it for the auxiliary diagnosis of ASD is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide an auxiliary diagnostic method and system for autism based on a two-person brain functional connectivity map and graph attention network, in order to solve the technical problems of existing ASD diagnosis relying on subjective behavioral assessment, lacking effective characterization of the neural mechanisms of two-way social interaction, and having low screening efficiency due to high professional threshold.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method and system for auxiliary diagnosis of autism based on a two-person brain functional connectivity graph and graph attention network includes a two-person fNIRS hyperscanning data acquisition module, a preprocessing and graph construction module, and an edge-weighted graph attention network classification module. The two-person fNIRS hyperscanning data acquisition module synchronously records the cortical hemodynamic activity of the subject and subject during a pre-set social interaction task, generating time-series data of synchronized brain activity. The preprocessing and graph construction module, connected to the two-person fNIRS hyperscanning data acquisition module, preprocesses the time-series data of synchronized brain activity and constructs a two-person brain functional connectivity graph containing nodes in the brain regions of the subject and subject. The edge-weighted graph attention network classification module, connected to the preprocessing and graph construction module, receives the two-person brain functional connectivity graph and dynamically aggregates and learns the node features and edge weights in the graph based on the graph attention mechanism, outputting autism classification results.

[0006] Preferably, the dual-person fNIRS ultrascanning data acquisition module includes a first fNIRS probe cap, a second fNIRS probe cap, a multi-channel fNIRS host, and an event marking unit; the first fNIRS probe cap is worn on the subject's head, and the second fNIRS probe cap is worn on the subject's head, with both probes covering the bilateral temporal, prefrontal, and parietal lobes; the multi-channel fNIRS host is connected to the first and second fNIRS probe caps respectively, with a sampling frequency of not less than 10Hz; the event marking unit is connected to the multi-channel fNIRS host and is used to record event markings for social interaction tasks.

[0007] Preferably, the preprocessing and graph construction module includes a signal preprocessing unit, a feature extraction unit, and a connectivity graph construction unit. The signal preprocessing unit performs motion artifact correction and 0.01-0.1Hz bandpass filtering on the raw fNIRS signal, and converts the optical density data into a time series of oxyhemoglobin and deoxyhemoglobin concentration changes. The feature extraction unit is connected to the signal preprocessing unit and is used to calculate the local features of each channel, including the sample entropy of oxyhemoglobin and deoxyhemoglobin, the coupling strength between them, and the average concentration change amplitude of each channel during the task. The connectivity graph construction unit is connected to the feature extraction unit and is used to calculate the interbrain functional connectivity strength between corresponding channel pairs of the experimenter and the subject using wavelet coherence methods, and construct a two-person brain functional connectivity graph.

[0008] Preferably, the edge-weighted enhanced graph attention network classification module includes an input layer, two graph attention layers, a global pooling layer, and a fully connected classification layer. The input layer receives the dual-brain functional connectivity graph and its node feature matrix. The two graph attention layers are connected to the input layer and are used to aggregate neighborhood information through an attention mechanism, wherein the calculation of the attention mechanism explicitly introduces and integrates the weight features of the connecting edges between nodes. The global pooling layer is connected to the graph attention layers and is used to aggregate node-level features into a graph-level representation. The fully connected classification layer is connected to the global pooling layer and is used to output auxiliary classification results for autism based on the graph-level representation.

[0009] Preferably, the preset social interaction tasks include eye contact, joint attention, and emotional feedback tasks.

[0010] Preferably, the time-series data of the synchronized brain activity of the two subjects includes data on changes in the concentrations of oxyhemoglobin and deoxyhemoglobin of the experimenter and the subject.

[0011] In the above technical solution, the present invention provides an auxiliary diagnostic method and system for autism based on a dual-brain functional connectivity map and graph attention network, which has the following beneficial effects: 1. This invention synchronously collects brain activity signals of the subject and the experimenter during social interaction through a dual-person fNIRS super-scanning data acquisition module, and constructs a dual-person brain functional connectivity map containing the brain region nodes and the strength of inter-brain connections of both parties. It can objectively quantify the abnormal inter-brain synchronization of ASD individuals in real social interactions, and effectively solves the problem of traditional diagnostic methods relying on subjective behavioral assessment and lacking objectivity. 2. This invention uses an edge-weighted enhanced graph attention network classification module to automatically extract and classify features from the functional connectivity graphs of two people's brains. It can quickly and accurately output auxiliary diagnostic results, effectively reduce the reliance on professional clinical experience in diagnosis, shorten the screening cycle, and provide quantifiable auxiliary screening tools for primary healthcare institutions. Attached Figure Description

[0012] Figure 1 This is a diagram of the architecture of the model based on an edge-weighted augmented graph attention network. Figure 2 This is a complete system flowchart. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0014] like Figure 1 and Figure 2 As shown, the present invention provides an autism auxiliary diagnosis method and system based on a dual-person brain functional connectivity graph and graph attention network, including a dual-person fNIRS superscan data acquisition module, a preprocessing and graph construction module, and an edge weight enhanced graph attention network classification module.

[0015] Example 1: System Overall Architecture and Workflow In this embodiment, refer to Figure 2 The complete system flowchart shown below details the overall workflow of the system. First, the hemodynamic activity of the cerebral cortex of the experimenter and subject during the social interaction task is simultaneously collected using a dual-person fNIRS super-scanning data acquisition module. Specifically, a standardized-trained experimenter (such as a child psychiatrist or a trained assessor) is paired with a subject to be assessed (a child suspected of having ASD or a child with typical developmental characteristics). Both wear a first fNIRS probe cap and a second fNIRS probe cap, respectively. The probe layout on the caps covers the bilateral temporal lobes, prefrontal lobes, and parietal lobes. These areas are key brain regions related to social cognitive function, including the superior temporal sulcus, temporoparietal junction, dorsolateral prefrontal lobe, and orbitofrontal lobe.

[0016] Before formal data collection, the system recorded event markers for the start, task switching, and end of the experiment using event marker units. The pre-set social interaction tasks included three blocks: an eye contact task (the experimenter and participant maintained continuous eye contact), a joint attention task (the experimenter guided the participant to focus on the same target object), and an emotional feedback task (the experimenter displayed different facial expressions, requiring the participant to identify and provide feedback). Each task block lasted approximately 3 minutes, with a 30-second resting state inserted between tasks as a baseline. The multi-channel fNIRS host synchronously recorded the near-infrared light intensity changes of both participants at a sampling frequency of 11Hz, forming the raw light intensity signal.

[0017] After acquisition, the raw data is transmitted to the preprocessing and graph construction module. The signal preprocessing unit first converts the raw light intensity signal into light density variation data. Then, it identifies anomalous abrupt changes in the signal using a motion artifact detection algorithm and corrects the identified artifacts using cubic spline interpolation. Next, a bandpass filter of 0.01-0.1 Hz is applied to the corrected data to eliminate physiological noise such as respiration (approximately 0.2-0.3 Hz) and heartbeat (approximately 1-1.5 Hz), as well as low-frequency signal drift. After filtering, based on a modified Beer-Lambert law, the light density data is converted into a time series of oxyhemoglobin (HbO) and deoxyhemoglobin (HbR) concentration changes.

[0018] The feature extraction unit receives the preprocessed HbO and HbR time series and calculates the local features for each channel. Taking the i-th channel as an example, it calculates the sample entropy value of the HbO signal, which reflects the signal complexity and typically ranges from 0 to 2. A higher entropy value indicates a more complex signal. It also calculates the Pearson correlation coefficient between HbO and HbR as the coupling strength between them, with a value ranging from -1 to 1. Positive values ​​indicate cooperative changes, while negative values ​​indicate antagonistic changes. Finally, it calculates the difference between the average HbO concentration during the task period (i.e., within the task block duration) and the baseline period as the activation amplitude feature.

[0019] Based on the aforementioned local features, the connectivity graph construction unit further calculates the strength of functional connections between brain regions. Taking the i-th channel of the experimenter and the j-th channel of the subject as an example, the synchronicity between the two is calculated using the wavelet coherence method. The wavelet coherence coefficient value is between 0 and 1, with the value closer to 1 indicating higher phase synchronicity of the channel pair in a specific frequency band. By calculating the coherence coefficients of all experimenter-subject channel pairs, a fully connected graph G containing 2N nodes (N being the number of channels on one side) is constructed, where each node corresponds to a channel, the node feature vector is the aforementioned extracted local feature, and the edge weights are the wavelet coherence coefficients.

[0020] Finally, the constructed dual-brain functional connectivity graph is input into the edge weighted graph attention network classification module. This module uses a pre-trained EWE-GAT model, which learns features from the graph structure data through two graph attention layers. After aggregation by a global pooling layer, the fully connected classification layer outputs the auxiliary classification results for autism (e.g., output probability values ​​greater than 0.5 indicate ASD tendency, otherwise it indicates TD tendency).

[0021] Example 2: Specific Implementation of the Dual-Person fNIRS Ultrascan Data Acquisition Module In this embodiment, the specific structure and working method of the dual-person fNIRS ultrascan data acquisition module are described in detail.

[0022] Reference Figure 1 and Figure 2 The dual-person fNIRS ultrascanning data acquisition module includes a first fNIRS probe cap, a second fNIRS probe cap, a multi-channel fNIRS host, and an event marking unit.

[0023] The first fNIRS probe cap is worn on the subject's head, and the second fNIRS probe cap is worn on the examiner's head; both have the same structure. Taking the first fNIRS probe cap as an example, it is made of elastic fabric and can be adjusted according to head circumference to ensure comfortable wear and good contact between the probe and the scalp. The probe layout adopts the international 10-10 system positioning standard, with a total of 12 light source probes and 8 detection probes, forming 20 measurement channels. Specifically, the layout is as follows: the frontal lobe region covers Fp1, Fp2, F3, and F4; the temporal lobe region covers T3, T4, T5, and T6; and the parietal lobe region covers P3, P4, and Pz. The light source probes emit near-infrared light with wavelengths of 730nm and 850nm, and the detection probes receive the emitted light intensity after tissue scattering.

[0024] The multi-channel fNIRS main unit uses fiber optic cables to connect to the first and second fNIRS probe caps respectively. The main unit has a built-in dual-channel synchronous acquisition card to ensure precise alignment of timestamps for both individuals' signals. The main unit's sampling frequency is set to 11Hz, which meets the Nyquist sampling requirements for hemodynamic signals (frequency below 0.5Hz). The main unit also features a real-time signal quality monitoring interface, allowing operators to view the signal strength of each channel in real time. If poor signal quality is detected in a particular channel (e.g., light intensity attenuation exceeding a threshold), the probe contact can be adjusted or the probe can be re-worn.

[0025] The event marking unit uses a button-type input device and connects to the multi-channel fNIRS host via a USB interface. During the experiment, the experimenter or operator presses the corresponding button at the beginning and end of each task block. The event marking unit writes the event code and timestamp to the data file for subsequent data segmentation.

[0026] In practical applications, the experimental environment was set in a quiet, well-lit room, with the experimenter and subject sitting opposite each other, approximately 1 meter apart. Before the experiment, both the experimenter and subject rested quietly for 5 minutes to allow their physiological states to stabilize. During the experiment, their behavior was recorded using a video monitoring system for subsequent joint analysis with fNIRS data.

[0027] Example 3: Specific Implementation of the Preprocessing and Graph Construction Module In this embodiment, the specific working method of the preprocessing and graph construction module is described in detail.

[0028] Reference Figure 1 and Figure 2 The preprocessing and graph construction module includes a signal preprocessing unit, a feature extraction unit, and a connection graph construction unit.

[0029] The signal preprocessing unit operates as follows: First, the raw light intensity signal is converted into light density changes. Then, a motion artifact detection algorithm based on amplitude and standard deviation thresholds is used to identify points in the signal where amplitude abrupt changes exceed three times the standard deviation. Cubic spline interpolation correction is then applied to the identified motion artifact intervals. Next, a Butterworth bandpass filter of 0.01–0.1 Hz is applied to the corrected data, with the order set to 4, to eliminate high-frequency noise such as breathing and heartbeat, as well as baseline drift. Finally, according to a modified Beer-Lambert law, the light density changes are converted into HbO and HbR concentration changes.

[0030] The feature extraction unit extracts features for each channel within the task block time window. Taking channel i of the experimenter as an example, the following features are extracted: HbO Sample Entropy: Using the sample entropy algorithm, with embedding dimension m=2 and similarity tolerance r=0.2×SD (SD is the signal standard deviation), the SampEn value is calculated, with a value range of 0-2.

[0031] HbR sample entropy: Calculated using the same method as above.

[0032] Coupling strength between HbO and HbR: Calculate the Pearson correlation coefficient r of the time series of HbO and HbR during the mission period.

[0033] Average HbO activation magnitude: Δ[HbO] is obtained by subtracting the baseline average from the average HbO concentration during the mission period.

[0034] HbR average activation magnitude: Calculated in the same way as above.

[0035] The connectivity graph construction unit calculates the strength of functional connections between brains. Taking the experimenter channel i and the subject channel j as an example, the synchronicity between the two is calculated using the wavelet coherence method. The specific steps are as follows: continuous wavelet transforms are performed on the two time series x(t) and y(t) to obtain wavelet coefficients. Finally, the average coherence coefficient in the 0.01-0.1Hz frequency band is taken as the edge weight. By traversing all experimenter-subject channel pairs, a two-person brain functional connectivity graph containing 2N nodes and N×N edges is constructed.

[0036] Example 4: Specific Implementation of the Edge-Weighted Graph Attention Network Classification Module In this embodiment, refer to Figure 1 The diagram shown illustrates the architecture of the edge-weighted augmented graph attention network, providing a detailed description of the specific structure and operation of the classification module.

[0037] The edge-weighted graph attention network classification module includes an input layer, two graph attention layers (first graph attention layer and second graph attention layer), a global pooling layer, and a fully connected classification layer.

[0038] The input layer receives the dual-brain functional connectivity graph G=(V,E) output by the preprocessing and graph construction modules, where V is the set of nodes, |V|=2N; E is the set of edges, and the edge weight is w. ij This represents the strength of the functional connectivity between node i and node j in the brain. Simultaneously, the input layer receives the node feature matrix X ∈ R. 2N×d , where d is the node feature dimension (in this embodiment, each node contains 5 features: HbO sample entropy, HbR sample entropy, HbO-HbR coupling strength, HbO average activation amplitude, and HbR average activation amplitude, so d=5).

[0039] The first graph attention layer performs feature transformation and attention aggregation on each node. Unlike traditional graph attention networks, this module explicitly introduces edge weights into the attention mechanism. Specifically, for node i and its neighbor node j, the attention coefficient e... ij The calculation formula is:

[0040] Where e ij,node e is obtained by multiplying the node features by the learnable attention vector after a linear transformation. ij,edge It is obtained by multiplying the edge features by the learnable attention vector after a linear transformation.

[0041] Then, the attention coefficients are normalized using the softmax function.

[0042] The first attention layer uses a multi-head attention mechanism, with 8 attention heads. Each head outputs a feature dimension d1=16. The results of the 8 heads are concatenated to obtain an output feature dimension of 128.

[0043] The second attention layer structure is similar to the first layer, but the output is a single head with a feature dimension d2=64. This layer further aggregates neighbor information to enhance the expressive power of the features.

[0044] The global pooling layer aggregates node-level features into a graph-level representation. It employs both global average pooling and global max pooling, averaging and maxing the features output from the second graph attention layer respectively. The results of these two pooling operations are then concatenated to obtain the graph-level feature vector g ∈ R. 128 (64-dimensional average pooling + 64-dimensional max pooling = 128-dimensional).

[0045] The fully connected classification layer consists of a Dropout layer (dropout rate of 0.5), a fully connected layer (128-dimensional input, 2-dimensional output), and a softmax activation function. The final output is a 2-dimensional probability vector, corresponding to the classification probabilities of "ASD propensity" and "TD propensity," respectively.

[0046] In practical applications, the model is pre-trained on a dataset containing 200 children with ASD and 200 children with TD, using the cross-entropy loss function and the Adam optimizer, with a learning rate of 0.001 and 200 training epochs. After training, the model parameters are fixed and used for auxiliary diagnosis of new samples.

[0047] Example 5: Practical Application Scenarios The workflow of this invention will be described below through a specific application scenario.

[0048] A five-year-old boy was brought to the pediatric psychiatry department of a top-tier hospital by his parents, complaining of difficulties in social interaction and avoidance of eye contact. The attending physician used the system of this invention for auxiliary assessment.

[0049] First, the physician (as the experimenter) and the child (as the subject) wore the first fNIRS probe cap and the second fNIRS probe cap, respectively. The system automatically performed signal quality checks, and after confirming that the signal quality of both individuals across 48 channels was good, the preset social interaction task program was initiated.

[0050] The task program ran for approximately 10 minutes and included three task blocks: eye contact, joint attention, and emotional feedback. During the experiment, the event tagging unit automatically recorded the start and end times of each task block. The multi-channel fNIRS host synchronously recorded the brain activity signals of both participants at an 11Hz sampling rate, and the data was saved to the local hard drive in real time.

[0051] After data acquisition, the preprocessing and graph construction module automatically processes the raw data: motion artifact correction, 0.01-0.1Hz bandpass filtering, conversion to HbO / HbR concentration change sequence, extraction of 5 local features of each channel, calculation of wavelet coherence coefficients of 20×20 pairs of channels as edge weights, and construction of a dual-brain functional connectivity graph.

[0052] The constructed graph data is input into the edge-weighted enhanced graph attention network classification module. After forward propagation, the model outputs: ASD propensity probability 0.87, TD propensity probability 0.13. The system displays the auxiliary diagnostic conclusion on the output interface: "ASD propensity, further evaluation with clinical scales is recommended."

[0053] Based on the ADOS-2 scale assessment results (total score 18, cutoff value 15) and clinical observation, the physician ultimately diagnosed ASD. The objective physiological data provided by this system offered strong neuroimaging evidence for the diagnosis, enhancing the physician's confidence in the diagnosis.

[0054] Example 6: Variant Implementation In another embodiment, the preset social interaction task of the present invention can be adjusted according to the assessment purpose. For example, for younger children (3-4 years old), the task can be simplified into a simple imitation game or a turn-taking game, and the task duration can be shortened to 2 minutes per block to maintain the child's attention. At the same time, the feature extraction unit can adjust the feature extraction window according to the task type, perform feature calculation only on the active period of the task, and eliminate invalid data segments.

[0055] In another implementation, the model parameters of the edge-weighted graph attention network classification module can be fine-tuned based on the data distribution of different centers. Primary healthcare institutions can use pre-trained models for direct inference, while large medical centers can collect local data to incrementally train the model, improving the classification accuracy of the model in specific populations.

[0056] In summary, the present invention provides an autism auxiliary diagnostic system based on a two-person brain functional connectivity map and a graph attention network. By using super-scanning technology to synchronously collect brain activity signals during two-person social interactions, a two-person brain functional connectivity map is constructed, and an edge weight-enhanced graph attention network is used for automatic classification. This achieves objective, rapid, and automated auxiliary diagnosis of ASD and has significant clinical application value.

[0057] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method and system for autism auxiliary diagnosis based on two-person brain functional connectivity graph and graph attention network, characterized in that, include: The dual-person fNIRS super-scan data acquisition module is used to synchronously record the cortical hemodynamic activity of the experimenter and the subject during a pre-set social interaction task, and generate time-series data of synchronous brain activity of the two people; The preprocessing and graph construction module is connected to the dual-person fNIRS super-scan data acquisition module and is used to preprocess the time series data of the dual-person synchronized brain activity and construct a dual-person brain functional connectivity graph containing brain region nodes of the experimenter and the subject. The edge weight-enhanced graph attention network classification module is connected to the preprocessing and graph construction module. It is used to receive the dual-brain functional connectivity graph, and dynamically aggregate and learn the node features and edge weights in the graph based on the graph attention mechanism, and output the autism classification result.

2. The autism auxiliary diagnosis method and system based on a dual-brain functional connectivity map and graph attention network according to claim 1, characterized in that, The dual-person fNIRS ultra-scan data acquisition module includes: The first fNIRS probe cap was worn on the subject's head, and its probe layout covered the bilateral temporal, prefrontal, and parietal lobe regions. The second fNIRS probe cap is worn on the subject's head, and its probe layout is the same as that of the first fNIRS probe cap. A multi-channel fNIRS host is connected to the first fNIRS probe cap and the second fNIRS probe cap respectively, for synchronous acquisition of fNIRS signals from two people; An event tagging unit, connected to the multi-channel fNIRS host, is used to record event tags for social interaction tasks.

3. The autism auxiliary diagnosis method and system based on a dual-brain functional connectivity map and graph attention network according to claim 2, characterized in that, The sampling frequency of the multi-channel fNIRS host is not less than 10Hz.

4. The autism auxiliary diagnosis method and system based on a dual-brain functional connectivity map and graph attention network according to claim 1, characterized in that, The preprocessing and graph construction module includes: The signal preprocessing unit is used to perform motion artifact correction and bandpass filtering on the raw fNIRS signal, and to convert the optical density data into a time series of changes in oxyhemoglobin and deoxyhemoglobin concentrations. The feature extraction unit, connected to the signal preprocessing unit, is used to calculate the local features of each channel respectively; The connectivity graph construction unit, connected to the feature extraction unit, is used to calculate the interbrain functional connectivity strength between corresponding channel pairs of the experimenter and the subject, and to construct a two-person brain functional connectivity graph.

5. The autism auxiliary diagnostic method and system based on a dual-brain functional connectivity map and graph attention network according to claim 4, characterized in that, The bandpass filter has a filtering range of 0.01-0.1Hz.

6. The autism auxiliary diagnostic method and system based on a dual-brain functional connectivity map and graph attention network according to claim 4, characterized in that, The local features include: the sample entropy of oxyhemoglobin and deoxyhemoglobin, the coupling strength between oxyhemoglobin and deoxyhemoglobin, and the average change in oxyhemoglobin and deoxyhemoglobin concentrations in each channel during the task.

7. The autism auxiliary diagnosis method and system based on a dual-brain functional connectivity map and graph attention network according to claim 4, characterized in that, The strength of the functional connectivity between brain cells was calculated using the wavelet coherence method.

8. The autism auxiliary diagnosis method and system based on a dual-brain functional connectivity map and graph attention network according to claim 1, characterized in that, The edge-weighted enhanced graph attention network classification module includes: The input layer is used to receive the dual-brain functional connectivity map and its node feature matrix; Two graph attention layers, connected to the input layer, are used to aggregate neighborhood information through an attention mechanism, wherein the calculation of the attention mechanism explicitly introduces and integrates the weight features of the connection edges between nodes; A global pooling layer, connected to the graph attention layer, is used to aggregate node-level features into a graph-level representation; A fully connected classification layer, connected to the global pooling layer, is used to output auxiliary classification results for autism based on the graph-level representation.

9. The autism auxiliary diagnostic method and system based on a dual-brain functional connectivity map and graph attention network according to claim 1, characterized in that, The preset social interaction tasks include eye contact, joint attention, and emotional feedback tasks.

10. The autism auxiliary diagnosis method and system based on a dual-brain functional connectivity map and graph attention network according to claim 1, characterized in that, The time-series data of the synchronized brain activity of the two subjects includes data on changes in the concentrations of oxyhemoglobin and deoxyhemoglobin of the experimenter and the subject.

Images