Multi-mode autism auxiliary diagnosis method and system based on EEG and ECG

Abstract

The invention discloses a multi-mode autism auxiliary diagnosis method and system based on EEG and ECG, and the system comprises a preprocessing module which is used for carrying out the preprocessing of a newly-tested EEG signal and a newly-tested ECG signal; the EEG coherent network construction module is used for calculating the coherent synchronism intensity of the electroencephalogram signals of any two electrodes based on the preprocessed EEG signals and constructing a whole-brain coherent network matrix; the EEG feature extraction module is used for extracting discriminative features based on the whole-brain coherent network matrix; the ECG feature extraction module is used for extracting heart rate variability features based on the preprocessed ECG signals; and the diagnosis module is used for inputting the discriminative features and the heart rate variability features into a pre-training classifier respectively, then carrying out weighted fusion on classification scores, and outputting a diagnosis result and a risk level according to the fusion scores. According to the method, objectivity, accuracy and grading reliability of ASD diagnosis can be improved, and a quantitative basis is provided for clinical early intervention.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of biomedical engineering and neuroengineering, and specifically relates to a multimodal autism auxiliary diagnostic method and system based on EEG and ECG. Background Technology

[0002] Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that begins in infancy or early childhood, characterized by impaired social communication, restricted interests, and repetitive behaviors, often accompanied by multiple comorbidities. The global prevalence of ASD exceeds 1% and continues to rise. Studies have shown that intervention before age 6 can lead to 2-3 times greater functional improvement in patients compared to untreated individuals, while missing this critical period may result in lifelong impairment. Therefore, early and accurate diagnosis is crucial for effective intervention.

[0003] Currently, the diagnosis of ASD mainly relies on the DSM-5 / ICD-11 behavioral standards and scales for assessment, but it has limitations such as strong subjectivity (the consistency among different assessors is only 60%-70%), dependence on professional resources (it is easy to miss or misdiagnose at the grassroots level), and low early identification rate (most are diagnosed after 2-3 years of age, far exceeding the optimal intervention period of 6-12 months).

[0004] To overcome these bottlenecks, researchers have begun actively exploring auxiliary diagnostic methods based on objective biomarkers. Electroencephalography (EEG), as a non-invasive and repeatable neurophysiological test, can reflect abnormal synchronicity in brain regions of patients with ASD (such as local hypersynchronization and weakened long-range connectivity). Heart rate variability (HRV) in electrocardiography (ECG) can effectively characterize autonomic nervous system dysregulation (such as decreased vagal tone). However, single-modality studies have limitations: EEG is easily affected by environmental and individual differences, with a cross-center consistency of 55%-65%; ECG HRV has insufficient specificity, overlaps with diseases such as ADHD, and has a misdiagnosis rate exceeding 20%. Furthermore, ASD involves dysregulation of central and autonomic nervous system interactions, and a single signal cannot fully reflect its pathological characteristics.

[0005] To fully utilize multimodal physiological signals, patent document CN116211306A discloses a mental health self-assessment system based on eye-tracking and electrocardiogram (ECG) signals. By combining ECG signal acquisition and analysis with eye-tracking signal acquisition and analysis, it analyzes the test subject's emotional stimulus responses. Utilizing a machine learning-based mental health assessment model, it can automatically identify mental illnesses and classify various conditions. However, it does not cover EEG, a key signal directly reflecting neurodevelopmental abnormalities, making it difficult to capture core biological characteristics such as abnormal brain functional connectivity in ASD patients. Patent document CN118506988A discloses a data processing method for autism rehabilitation training, integrating multimodal signals such as ECG, EEG, speech, and eye-tracking for real-time emotion recognition and adaptive adjustment in rehabilitation training. However, its focus is on emotion recognition and adaptive adjustment in rehabilitation training, and it lacks sufficient exploration and validation of early-stage ASD-specific biomarkers.

[0006] Therefore, how to more effectively integrate EEG and ECG multimodal physiological signals and build an early identification and diagnosis system that combines objectivity and accuracy has become the key to breaking through technical bottlenecks, which can improve diagnostic accuracy and promote the transformation to precision driven by biological mechanisms. Summary of the Invention

[0007] In view of the above, the purpose of this invention is to provide a multimodal autism auxiliary diagnostic method and system based on EEG and ECG, which overcomes the limitations of existing single-modal diagnostic technologies. By integrating the discriminative features of the EEG whole-brain coherent network matrix and the ECG heart rate variability features, the objectivity, accuracy and grading reliability of ASD diagnosis are improved, providing quantitative evidence for early clinical intervention.

[0008] To achieve the above-mentioned objectives, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide a multimodal autism auxiliary diagnostic system based on EEG and ECG, comprising: The preprocessing module is used to perform preprocessing on the EEG and ECG signals of the new subjects; The EEG coherent network construction module is used to calculate the synchronization strength of the coherence of EEG signals between any two electrodes based on the preprocessed EEG signals and to construct a whole-brain coherent network matrix. The EEG feature extraction module is used to extract discriminative features based on the whole-brain coherent network matrix; The ECG feature extraction module is used to extract heart rate variability features based on the preprocessed ECG signal; The diagnostic module is used to input discriminative features and heart rate variability features into a pre-trained classifier, then weight and fuse the classification scores, and output the diagnostic results and risk level based on the fused score.

[0009] Preferably, in the preprocessing module, the preprocessing process for the EEG signal includes rereference correction, bandpass filtering for noise reduction, independent component analysis for artifact removal, and segmented signal generation; the preprocessing process for the ECG signal includes filtering, R-wave peak detection, and RR interval sequence generation.

[0010] Preferably, in the EEG coherence network construction module, the synchronization strength of the coherence between any two electrode EEG signals is calculated based on the preprocessed EEG signals, and a whole-brain coherence network matrix is ​​constructed, including: For each segment of the preprocessed EEG signal Several window signals are obtained by performing sliding windowing processing. , Indicates a segmented index. Indicates time, Indicates the window index; For each window signal Any two electrode channels within and Calculate its amplitude squared coherence coefficient Used to reflect the frequency of the two channel signals. Synchronization level below: , in, and They represent channels respectively. and channels The power spectral density, This represents the cross-power spectral density of the two channels; The mean of the squared amplitude coherence coefficients at all frequency points within each segment is taken as the inter-channel coherence value. : , in, Indicates the first segment within the segment Each frequency point, This indicates the number of frequency points contained in the segment; Combine the inter-channel coherence values ​​of all channel pairs within each segment into a symmetric matrix. Used to reflect the strength of functional connectivity within brain segments: , in, Indicates the number of channels; Symmetric matrix for all segments Averaging yields the final whole-brain coherent network matrix. Used to reflect the strength of functional connectivity throughout the brain: , in, This indicates the total number of segments.

[0011] Preferably, in the EEG feature extraction module, discriminative features are extracted based on the whole-brain coherent network matrix, including: The obtained whole-brain coherent network matrix is ​​projected onto a pre-trained multi-pair spatial filter to obtain the projection result. The variance of each row of the projection result is calculated, and the variance is logarithmically normalized to finally obtain the multidimensional EEG discriminative features.

[0012] Preferably, in the ECG feature extraction module, heart rate variability features are extracted based on the preprocessed ECG signal, including: Heart rate variability features were calculated based on the RR interval sequence of the preprocessed ECG signal, including: mean RR interval, standard deviation of RR interval, root mean square of the difference between adjacent RR intervals, minimum RR interval, maximum RR interval, mean heart rate, standard deviation of heart rate, high-frequency / low-frequency power ratio, sample entropy, short axis of Poincaré plot, and long axis of Poincaré plot.

[0013] Preferably, in the diagnostic module, after inputting the discriminative features and heart rate variability features into a pre-trained classifier, the classification scores are weighted and fused, including: The discriminative features of the EEG signal are input into the first classifier to obtain the first decision value, and the heart rate variability features of the ECG signal are input into the second classifier to obtain the second decision value. The first and second decision values ​​are respectively subjected to PlattScaling probability calibration, and the calibrated probability values ​​are weighted and summed according to the weight coefficients to obtain the fusion score.

[0014] Preferably, in the diagnostic module, the diagnostic result and risk level are output based on the fusion score, including: The diagnosis of autism is determined based on the fusion score and the optimal classification threshold, and the risk level is determined based on the fusion score and the risk quantile.

[0015] Secondly, embodiments of the present invention also provide a multimodal autism auxiliary diagnosis method based on EEG and ECG, implemented based on the aforementioned multimodal autism auxiliary diagnosis system based on EEG and ECG, comprising the following steps: The preprocessing module was used to preprocess the EEG and ECG signals of the new subjects; The EEG coherence network construction module is used to calculate the synchronicity strength of coherence between any two electrodes of EEG signals based on preprocessed EEG signals and to construct a whole-brain coherence network matrix. Discriminative features were extracted based on the whole-brain coherent network matrix using the EEG feature extraction module; Heart rate variability features were extracted from the preprocessed ECG signal using the ECG feature extraction module. The diagnostic module inputs discriminative features and heart rate variability features into a pre-trained classifier, then weights and fuses the classification scores, and outputs the diagnostic results and risk levels based on the fused scores.

[0016] Thirdly, embodiments of the present invention also provide an electronic device, including a memory and one or more processors, wherein the memory is used to store a computer program, and the processor is used to implement the above-described multimodal autism auxiliary diagnosis method based on EEG and ECG when executing the computer program.

[0017] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, implements the above-described multimodal autism auxiliary diagnosis method based on EEG and ECG.

[0018] Compared with the prior art, the beneficial effects of the present invention include at least the following: This invention utilizes EEG and ECG multimodal fusion technology to achieve accurate diagnosis and risk stratification of ASD. Specifically, the EEG coherence network construction module and EEG feature extraction module construct a whole-brain coherence network and extract discriminative features, accurately capturing functional connectivity abnormalities in the central nervous system of ASD patients. Furthermore, combined with the ECG feature extraction module, multidimensional heart rate variability features are extracted from RR interval sequences for multimodal fusion analysis, revealing the complex characteristics of the interaction and regulation between the central nervous system and the autonomic nervous system, successfully elucidating the multidimensional pathological mechanisms of ASD patients at the neuro-autonomic level. Compared to traditional behavioral scale assessment methods, EEG and ECG, as physiological signals, can more objectively and directly capture abnormal changes in brain neural activity and autonomic nervous function, thereby enabling the discovery of highly specific biomarkers. Moreover, the application of resting-state EEG and conventional ECG simplifies data collection and avoids diagnostic errors caused by the assessor's subjective experience. In summary, this invention breaks away from the framework of traditional behavioral assessment and uses objective multimodal physiological characteristics to achieve accurate diagnosis and classification of ASD, providing reliable and easy-to-use technical support for large-scale early screening and personalized intervention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the structure of the multimodal autism auxiliary diagnostic system based on EEG and ECG provided in an embodiment of the present invention; Figure 2 This is a box plot of risk grading results provided in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the multimodal autism auxiliary diagnosis method based on EEG and ECG provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of this invention.

[0022] The inventive concept of this invention is as follows: Addressing the problems in existing autism diagnosis technologies, such as reliance on subjective behavioral scales, lack of objective biomarkers, and the inability of single-modal physiological signals to comprehensively reflect pathological mechanisms, this invention provides a multimodal autism auxiliary diagnostic method and system based on EEG and ECG. The core of this method lies in: constructing a whole-brain coherent network from the collected EEG signals to quantify brain functional connectivity; and simultaneously extracting multidimensional heart rate variability features from ECG signals to assess autonomic nervous system status. This approach more objectively and directly captures abnormalities in brain neural activity and autonomic nervous function, facilitating the discovery of highly specific biomarkers. Furthermore, a classifier network is used to effectively fuse and make decisions regarding the aforementioned multimodal biomarkers, ultimately outputting a comprehensive assessment that combines binary diagnostic results with risk levels. This solution breaks through the framework of traditional behavioral assessments, providing a quantifiable and easy-to-use objective diagnostic tool for early clinical diagnosis.

[0023] like Figure 1 As shown, the embodiment provides a multimodal autism auxiliary diagnostic system 100 based on EEG and ECG, including: a preprocessing module 110, an EEG coherent network construction module 120, an EEG feature extraction module 130, an ECG feature extraction module 140, and a diagnostic module 150.

[0024] (1) The preprocessing module 110 is used to perform preprocessing on the EEG and ECG signals of the new subjects.

[0025] In this embodiment, the preprocessing procedure for the EEG signal is as follows: Step 1: Average and recalibrate the original EEG signal to eliminate potential deviation caused by a single reference electrode.

[0026] Step 2: Use a bandpass filter to filter the corrected EEG signal. The preferred filtering range is 0.5–45Hz to remove DC drift, power frequency interference and high-frequency electromyographic noise, thereby preserving the main frequency band components of brain electrical activity.

[0027] Step 3: Perform independent component analysis on the filtered EEG signal to identify and remove artifacts related to blinking, electromyography, or movement, thereby obtaining a pure neural activity signal.

[0028] Step 4: Divide the cleaned continuous EEG signal into fixed-length time segments, each segment lasting 5 seconds, with a sampling rate of [missing information]. =512Hz, each segment contains 5*512=2560 sampling points, let the segmented signal be represented as: , in, Indicates the number of segments. Indicates the first A 5-second signal segment, Indicates time.

[0029] In this embodiment, the preprocessing procedure for the ECG signal is as follows: Step 1: Use a 0.5-40Hz bandpass filter to remove baseline drift and high-frequency noise, and use a notch filter to eliminate high-frequency interference and harmonics.

[0030] Step 2: Use the Pan-Tompkins algorithm to perform R-wave detection on the filtered signal.

[0031] Step 3: Calculate the time interval between adjacent R peaks based on the detected R peak positions to obtain the RR intervals. Apply a physiological rationality filter to remove abnormal intervals and generate the final RR interval sequence for subsequent analysis.

[0032] (2) The EEG coherence network construction module 120 is used to calculate the synchronization intensity of the coherence of the EEG signals of any two electrodes based on the preprocessed EEG signals and to construct the whole brain coherence network matrix.

[0033] In this embodiment, the process of constructing the whole-brain coherent network matrix is ​​as follows: Step 1: Process each segment of the preprocessed EEG signal Within each segment, a Hamming window function is used to perform sliding windowing processing on the signal. The window length is set to 256 sampling points, and adjacent windows overlap by 128 sampling points (overlap rate 50%) to ensure signal continuity and reduce edge effects introduced by segmentation. After windowing processing, the signal... Each segment is divided into several window signals. ,in Indicates the window index.

[0034] Step 2: For each window signal Any two electrode channels within and Calculate its amplitude squared coherence coefficient Used to reflect the frequency of the two channel signals. Synchronization level below: , in, and They represent channels respectively. and channels The power spectral density, The cross-power spectral density of the two channels is represented by the Fast Fourier Transform (FFT, with 512 transform points) and the spectral resolution is 1 Hz. The value range is from 0 to 1.

[0035] Step 3: Take the mean of the squared amplitude coherence coefficients of all frequency points within each segment (theta band, 4-8Hz) as the inter-channel coherence value. : , in, Indicates the first segment within the segment Each frequency point, This indicates the number of frequency points contained in the segment.

[0036] Step 4: Combine the inter-channel coherence values ​​of all channel pairs within each segment into a symmetric matrix. Used to reflect the strength of functional connectivity within brain segments: , in, Indicates the number of channels.

[0037] Step 5: For all segmented symmetric matrices Averaging yields the final whole-brain coherent network matrix. Used to reflect the strength of functional connectivity throughout the brain: , in, This indicates the total number of segments.

[0038] (3) The EEG feature extraction module 130 is used to extract discriminative features based on the whole brain coherent network matrix.

[0039] In this embodiment, the process for extracting discriminative features is as follows: Step 1: Process the extracted whole-brain coherent network matrix Using the CSP filtering parameters determined during the pre-training phase, five pairs of preset filters are invoked, and the filter matrix is ​​represented as follows: : .

[0040] Step 2: Integrate the whole-brain coherent network matrix Projecting the image onto a preset filter yields the projection result. : .

[0041] Step 3: Calculation variance of each row 10-dimensional discriminative features are obtained through logarithmic normalization: , This discriminative feature will be used as input to a subsequent classifier to predict the subject's category.

[0042] (4) The ECG feature extraction module 140 is used to extract heart rate variability features based on the preprocessed ECG signal.

[0043] In this embodiment, the process for extracting heart rate variability features is as follows: Step 1: Extract the following features based on the RR interval sequences generated after preprocessing: Mean RR interval: ,in Indicates the index of the RR interval sequence. Indicates the number of RR interval sequences; RR interval standard deviation (SDNN): ; Root mean square of the difference between adjacent RR intervals (RMSSD): ; Minimum RR interval: ; Maximum RR interval: ; Mean heart rate: ,in ; Heart rate standard deviation: .

[0044] Step 2: Calculate the low-frequency / high-frequency power ratio (LF / HF) using power spectrum analysis: Define the low-frequency band (LF) as 0.04-0.15 Hz and the high-frequency band (HF) as 0.15-0.4 Hz, and calculate the power spectrum integral values ​​for both frequency bands. and but: .

[0045] Step 3: Extract sample entropy features and construct... dimensional vector sequence ( ), define the distance between vectors Statistical distance less than the threshold ( Standard deviation of RR interval series The vector pair ratio is 0.1-0.2 times that of the vector pair. Calculate the sample entropy: .

[0046] Step 4: Calculate the minor axis of the Poincaré plot (SD1): Poincaré diagram, long axis (SD2): .

[0047] (5) The diagnostic module 150 is used to input the discriminative features and heart rate variability features into the pre-trained classifier, then weight and fuse the classification scores, and output the diagnostic results and risk level based on the fused score.

[0048] In this embodiment, the CSP filtering parameters required by the EEG feature extraction module 130 are first pre-trained in the diagnostic module 150, and the process is as follows: Step 1: Obtain a training dataset labeled with ASD / TD (Autism Spectrum Disorder / Typical Development), which includes raw EEG and ECG signals from several subjects. Perform rereference correction, bandpass filtering for noise reduction, independent component analysis for artifact removal, and 5-second segmented signal generation on the EEG signals according to the preprocessing module 110. Then, calculate the whole-brain EEG coherence network matrix for each subject according to the process described in the EEG coherence network construction module 120. .

[0049] Step 2: Determine the CSP filter parameters: Step 2.1: Press the label The whole-brain coherent network matrix of the training data Collections classified into TD classes A collection of ASD classes ; Step 2.2: Category Each whole-brain coherent network matrix Calculate its covariance matrix superscript This represents the transpose and normalization to a matrix. , Representing the trace of the matrix yields the category. Mean covariance matrix ,in Indicates category The number of training samples; Step 2.3: Calculate the average covariance matrix of the TD class. The average covariance matrix of the ASD class Combined covariance matrix Perform feature decomposition Then, a whitening matrix is ​​constructed. ,satisfy ( (the identity matrix); Step 2.4: Analyze the whitened covariance matrix Perform generalized feature decomposition: The eigenvector matrix is ​​obtained. ; Step 2.5: Construct the spatial filter matrix Five pairs of filters with the strongest discriminative power (corresponding to the eigenvectors of the five largest and five smallest eigenvalues) are selected to form a preset filter matrix: , as CSP filter parameters, where The eigenvectors corresponding to the five largest eigenvalues ​​in the generalized eigenvalue decomposition. The eigenvector corresponding to the minimum 5 eigenvalues.

[0050] In this embodiment, two linear kernel function SVM classifiers (the first classifier for EEG) are pre-trained in the diagnostic module 150. The second classifier with ECG The process is as follows: Step 1: Obtain a training dataset labeled with ASD / TD (Autism Spectrum Disorder / Typical Development), which contains raw EEG and ECG signals from several subjects. Extract 10-dimensional discriminative features and 11-dimensional heart rate variability features from the subject samples in the training dataset according to the workflow of EEG feature extraction module 130 and ECG feature extraction module 140.

[0051] Step 2: Perform max-min standardization (scaling to the [0,1] interval) on both features respectively. Step 3: Construct a linear kernel SVM classifier based on the discriminative features of the standardized EEG signal. The optimal parameters are obtained by solving the following optimization problem: , , in, Indicates the first EEG discriminant characteristics of a sample of subjects For the corresponding tags, TD is 1 and ASD is 2. and Parameters (weights and biases) for the classification hyperplane. The penalty factor (determined via grid search) is used. As slack variables, This represents the total number of samples in the training dataset. Step 4: Based on the heart rate variability characteristics of the standardized ECG signal, train the ECG classifier using the same linear kernel SVM model as in the previous step. The optimal parameters (weights) are obtained through optimization. Bias and punishment factors ); Step 5: [Regarding...] and The performance was evaluated using five-fold cross-validation to ensure the model's generalization ability.

[0052] In this embodiment, a graded diagnosis is finally performed based on the two pre-trained classifiers, and the process is as follows: Step 1: Based on the samples in the training dataset, perform the first classifier for EEG. The first decision value output Construct a Platt Scaling probability calibration model to convert decision values ​​into probabilities in the [0,1] interval: ,in and For calibration parameters (solved through maximum likelihood estimation); similarly, for the second classifier of ECG The second decision value output Constructing a Platt Scaling probability calibration model: ,in and For calibration parameters; Step 2: Obtain the raw first decision values ​​of the EEG from the samples in the training dataset. ECG original second decision value Substituting into the calibration model, we obtain the calibrated probability values: Based on the AUC curve performance of the two classifiers, the weighting coefficients are used. and Perform weighted fusion to generate a training set fusion score: ; Step 3: Plot the ROC curve by combining the training fusion score and the true label, and determine the optimal classification threshold using the Youden exponent maximization criterion. The value is 0.272. At the same time, the key quantiles of the training fusion score (25%, 50%, 75%, 90%) are calculated and are 0.10, 0.13, 0.69, and 0.76, respectively, serving as the dividing points for risk levels (low risk, low-to-medium risk, medium-to-high risk, high risk, and extremely high risk). Step 4: Perform the following diagnostic procedure on the new subject samples: Step 4.1: Extract the 10-dimensional EEG discriminative features and 11-dimensional ECG heart rate variability features from the new subject samples, and input them into the classifier after standardization. To obtain the original decision value

[0053] 4.2 Applying the calibration model to calculate probability values: ; 4.3 Calculate the fusion score: ; 4.4 Classification based on threshold: If It is determined to be ASD, otherwise it is TD; at the same time, the training set quantiles (< For low risk, And < For medium and low risk, And < For medium and high risk, And < For high risk, (Assuming extremely high risk), the final output will be a diagnosis result of "category + risk level".

[0054] To verify the effectiveness of this invention, the trained EEG and ECG-based multimodal autism auxiliary diagnostic system was applied to the EEG and ECG data of 55 subjects (18 ASD patients and 37 TD children), and performance was evaluated using five-fold cross-validation. Specifically, 90% of the samples from the 55 subjects were selected for model generation and evaluation, and the remaining 10% were used as an independent diagnostic test set. The risk grading results based on the fusion score of EEG coherence features and ECG heart rate variability features are as follows: Figure 2 As shown in the results, the fusion model proposed in this invention (i.e., weighted fusion after scoring by two classifiers) can effectively distinguish between ASD patients and TD children, indicating that the features extracted in this invention have practical application value in ASD diagnosis and classification, and lay the foundation for achieving multimodal fusion to improve the diagnosis and classification effect.

[0055] Based on the same inventive concept, such as Figure 3 As shown in the figure, this embodiment of the invention also provides a multimodal autism-assisted diagnostic method based on EEG and ECG, including the following steps: S1, the preprocessing module is used to preprocess the EEG and ECG signals of the new subjects.

[0056] S2 utilizes the EEG coherence network construction module to calculate the synchronization intensity of coherence between any two-electrode EEG signals based on the preprocessed EEG signals and constructs a whole-brain coherence network matrix.

[0057] S3 utilizes the EEG feature extraction module to extract discriminative features based on the whole-brain coherent network matrix.

[0058] S4. The ECG feature extraction module is used to extract heart rate variability features based on the preprocessed ECG signal.

[0059] S5 uses the diagnostic module to input discriminative features and heart rate variability features into the pre-trained classifier, then weights and fuses the classification scores, and outputs the diagnostic results and risk level based on the fused score.

[0060] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, including a memory and one or more processors, wherein the memory is used to store a computer program, and the processor is used to implement the above-described multimodal autism auxiliary diagnosis method based on EEG and ECG when executing the computer program.

[0061] Based on the same inventive concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, implements the above-described multimodal autism auxiliary diagnosis method based on EEG and ECG.

[0062] It should be noted that the multimodal autism auxiliary diagnosis method, electronic device, and computer-readable storage medium based on EEG and ECG provided in the above embodiments all belong to the same inventive concept as the multimodal autism auxiliary diagnosis system based on EEG and ECG. For details of their specific implementation process, please refer to the embodiments of the multimodal autism auxiliary diagnosis system based on EEG and ECG, which will not be repeated here.

[0063] The specific embodiments described above illustrate the technical solution and beneficial effects of the present invention in detail. It should be understood that the above description is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multimodal autism auxiliary diagnostic system based on EEG and ECG, characterized in that, include: The system includes a preprocessing module, an EEG coherent network construction module, an EEG feature extraction module, an ECG feature extraction module, and a diagnostic module. The preprocessing module is used to perform preprocessing on the EEG and ECG signals of the new subjects; The EEG coherent network construction module is used to calculate the synchronization strength of the coherence of EEG signals between any two electrodes based on the preprocessed EEG signals and to construct a whole-brain coherent network matrix. The EEG feature extraction module is used to extract discriminative features based on the whole-brain coherent network matrix; The ECG feature extraction module is used to extract heart rate variability features based on the preprocessed ECG signal; The diagnostic module is used to input discriminative features and heart rate variability features into a pre-trained classifier, then weight and fuse the classification scores, and output the diagnostic results and risk level based on the fused score.

2. The multimodal autism auxiliary diagnostic system based on EEG and ECG according to claim 1, characterized in that, In the preprocessing module, the preprocessing flow for EEG signals includes rereference correction, bandpass filtering for noise reduction, independent component analysis for artifact removal, and segmented signal generation; the preprocessing flow for ECG signals includes filtering, R-wave peak detection, and RR interval sequence generation.

3. The multimodal autism auxiliary diagnostic system based on EEG and ECG according to claim 1, characterized in that, In the EEG coherence network construction module, the synchronization intensity of coherence between any two-electrode EEG signals is calculated based on the preprocessed EEG signals, and a whole-brain coherence network matrix is ​​constructed, including: For each segment of the preprocessed EEG signal Several window signals are obtained by performing sliding windowing processing. , Indicates a segmented index. Indicates time, Indicates the window index; For each window signal Any two electrode channels within and Calculate its amplitude squared coherence coefficient Used to reflect the frequency of the two channel signals. Synchronization level below: ， in, and Representing channels and channels The power spectral density, This represents the cross-power spectral density of the two channels; The mean of the squared amplitude coherence coefficients at all frequency points within each segment is taken as the inter-channel coherence value. : ， in, Indicates the first segment within the segment Each frequency point, This indicates the number of frequency points contained in the segment; Combine the inter-channel coherence values ​​of all channel pairs within each segment into a symmetric matrix. Used to reflect the strength of functional connectivity within brain segments: ， in, Indicates the number of channels; Symmetric matrix for all segments Averaging yields the final whole-brain coherent network matrix. Used to reflect the strength of functional connectivity throughout the brain: ， in, This indicates the total number of segments.

4. The multimodal autism auxiliary diagnostic system based on EEG and ECG according to claim 1, characterized in that, In the EEG feature extraction module, discriminative features are extracted based on the whole-brain coherent network matrix, including: The obtained whole-brain coherent network matrix is ​​projected onto a pre-trained multi-pair spatial filter to obtain the projection result. The variance of each row of the projection result is calculated, and the variance is logarithmically normalized to finally obtain the multidimensional EEG discriminative features.

5. The multimodal autism auxiliary diagnostic system based on EEG and ECG according to claim 1, characterized in that, In the ECG feature extraction module, heart rate variability features are extracted based on the preprocessed ECG signal, including: Heart rate variability features were calculated based on the RR interval sequence of the preprocessed ECG signal, including: mean RR interval, standard deviation of RR interval, root mean square of the difference between adjacent RR intervals, minimum RR interval, maximum RR interval, mean heart rate, standard deviation of heart rate, high-frequency / low-frequency power ratio, sample entropy, short axis of Poincaré plot, and long axis of Poincaré plot.

6. The multimodal autism auxiliary diagnostic system based on EEG and ECG according to claim 1, characterized in that, In the diagnostic module, after inputting discriminative features and heart rate variability features into a pre-trained classifier, the classification scores are weighted and fused, including: The discriminative features of the EEG signal are input into the first classifier to obtain the first decision value, and the heart rate variability features of the ECG signal are input into the second classifier to obtain the second decision value. The first and second decision values ​​are respectively subjected to Platt Scaling probability calibration, and the calibrated probability values ​​are weighted and summed according to the weight coefficients to obtain the fusion score.

7. The multimodal autism auxiliary diagnostic system based on EEG and ECG according to claim 1, characterized in that, The diagnostic module outputs diagnostic results and risk levels based on the fusion score, including: The diagnosis of autism is determined based on the fusion score and the optimal classification threshold, and the risk level is determined based on the fusion score and the risk quantile.

8. A multimodal autism auxiliary diagnosis method based on EEG and ECG, implemented based on the multimodal autism auxiliary diagnosis system based on EEG and ECG as described in any one of claims 1 to 7, characterized in that, Includes the following steps: The preprocessing module was used to preprocess the EEG and ECG signals of the new subjects; The EEG coherence network construction module is used to calculate the synchronicity strength of coherence between any two electrodes of EEG signals based on preprocessed EEG signals and to construct a whole-brain coherence network matrix. Discriminative features were extracted based on the whole-brain coherent network matrix using the EEG feature extraction module; Heart rate variability features were extracted from the preprocessed ECG signal using the ECG feature extraction module. The diagnostic module inputs discriminative features and heart rate variability features into a pre-trained classifier, then weights and fuses the classification scores, and outputs the diagnostic results and risk levels based on the fused scores.

9. An electronic device comprising a memory and one or more processors, the memory for storing a computer program, characterized in that, The processor is used to implement the multimodal autism auxiliary diagnosis method based on EEG and ECG as described in claim 8 when executing a computer program.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by a computer, it implements the multimodal autism auxiliary diagnosis method based on EEG and ECG as described in claim 8.

Images