A multi-modal neuro-physiological signal processing method, system and readable storage medium

Abstract

The application discloses a multi-modal neurophysiological signal processing method, system and readable storage medium. The method comprises the following steps: passively collecting multi-modal physiological signals of a subject in a resting state and an induced task state; acquiring the cooperation degree features and signal quality features of the subject in the induced task state, and dynamically adjusting the induced task content and presentation rhythm by using a language model; performing source positioning reverse solving on the electroencephalogram signals based on a template skull model of a corresponding age stage to extract source space features; calculating the deviation amount features of the source space features relative to development norm data and fusing to generate a multi-modal development deviation feature vector; inputting the multi-modal development deviation feature vector into a deviation evaluation model for processing, and outputting state deviation score data and a data analysis report. The application can reduce the source positioning error caused by physical changes with age, strip the development fluctuation interference, and improve the signal quality.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and neural signal processing technology, and more particularly to a multimodal neurophysiological signal processing method, system, and readable storage medium. Specifically, this application covers the application of multimodal data fusion, machine learning classification, physiological signal feature extraction, and computer-aided assessment systems in medical and health informatics. Background Technology

[0002] The need for objective assessment of the emotional and cognitive states of specific age groups is increasingly prominent. Traditional assessment methods mainly rely on subjective scales, which are highly subjective and lack objective biological indicators, resulting in poor stability and repeatability of assessment results, hindering large-scale deployment. Existing objective detection technologies typically collect electroencephalogram (EEG) or functional magnetic resonance imaging (fMRI) signals and directly extract features at the sensor level for classification. However, most existing schemes ignore the physical fact that skull thickness and conductivity change rapidly with age in developing individuals, leading to spatial drift and attenuation errors in brain power localization results based on electromagnetic field inverse solving. Furthermore, due to the low cooperation and high head movement frequency of underage subjects, fixed evoked task paradigms result in a large number of motion artifacts in the collected physiological signals, reducing the signal-to-noise ratio. In addition, existing multimodal fusion schemes fail to isolate the interference of natural developmental background fluctuations caused by age, resulting in insufficient generalization ability. To address the aforementioned objective engineering and technical deficiencies, it is essential to provide a multimodal neurophysiological signal processing scheme. Summary of the Invention

[0003] The purpose of this application is to provide a multimodal neurophysiological signal processing method, system, and readable storage medium to solve the problems in the prior art that lead to brain power source localization errors due to ignoring changes in the physical characteristics of the skull during the development of the subject, and poor signal quality caused by fixed task paradigms.

[0004] To achieve the above objectives, a first aspect of this application provides a multimodal neurophysiological signal processing method, comprising:

[0005] Multimodal physiological signals of subjects were passively collected in both resting and induced task states; wherein the multimodal physiological signals included at least electroencephalogram (EEG) signals.

[0006] In the induced task state, the cooperation characteristics and signal quality characteristics of the subject are obtained, and the induced task content and presentation rhythm provided to the subject are dynamically adjusted according to the cooperation characteristics and signal quality characteristics using a preset language model;

[0007] The collected EEG signals are used to perform source localization inverse solution based on the template skull model of the subject's corresponding age group to extract the source spatial features of the corresponding brain regions.

[0008] The developmental norm data corresponding to the multimodal physiological signals are obtained, the deviation features of the source spatial features relative to the developmental norm data are calculated, and the deviation features are fused to generate a multimodal developmental deviation feature vector.

[0009] The multimodal developmental deviation feature vector is input into a preset deviation assessment model for processing, and the corresponding state deviation score data and data analysis report are output.

[0010] The steps of source localization and reverse solution based on the template skull model corresponding to the subject's age group include: obtaining the subject's actual age information; retrieving the template skull model that matches the actual age information from a preset model database; wherein, the template skull model includes skull thickness parameters and electrical conductivity parameters corresponding to a specific age development stage.

[0011] The template skull model is configured as a skull boundary element model; the step of performing source localization inverse solution based on the template skull model corresponding to the subject's age group specifically includes: substituting the skull thickness parameter and the conductivity parameter into the skull boundary element model to perform forward electromagnetic field modeling; and using a precise low-resolution EEG to solve the EEG signal to obtain the current density distribution of the scalp signal in the cortical source space.

[0012] The precise low-resolution EEG tomography algorithm performs the following operations: acquiring a three-dimensional spatial coordinate system including scalp electrode location information and the skull boundary element model; constructing a positive conduction lead field matrix based on the conductivity parameters, the positive conduction lead field matrix reflecting the electromagnetic attenuation mapping relationship from the source space of the cerebral cortex to the scalp surface; introducing a spatial smoothing prior assumption to calculate the electrode signal covariance matrix of the source space; calculating a zero-error source localization weight matrix based on the positive conduction lead field matrix and the electrode signal covariance matrix; multiplying the continuous time series of the acquired EEG signals with the zero-error source localization weight matrix to output the three-dimensional dipole moment time series corresponding to the three-dimensional voxels in the source space, and using the frequency band energy of the three-dimensional dipole moment time series as the source space feature.

[0013] The steps of obtaining the subject's cooperation characteristics and signal quality characteristics include: obtaining the subject's gaze disengagement rate through an eye-tracking device; obtaining the subject's head movement frequency through a posture sensor; and combining the gaze disengagement rate and the head movement frequency to form the cooperation characteristics.

[0014] The step of dynamically adjusting the induced task content and presentation rhythm provided to the subject based on the cooperation characteristics and signal quality characteristics using a preset language model includes: using the language model to perform time series prediction of the gaze disengagement rate and the head movement frequency to generate a fatigue trend index; and issuing control instructions based on the fatigue trend index to control the task presentation terminal to reduce the visual complexity of the induced task content and extend the interval time of the presentation rhythm.

[0015] The preset language model performs the following operations: establishing a continuous state memory sliding window containing historical task execution records, the current gaze disengagement rate time series, and the current head movement frequency; extracting the state deterioration gradient value within the continuous state memory sliding window; triggering a dynamic intervention mechanism when the state deterioration gradient value exceeds a preset fatigue threshold; the dynamic intervention mechanism generates a reconstructed task flow sequence containing task element simplification instructions, stimulus presentation duration extension instructions, and positive visual feedback elements by inputting the state deterioration gradient value into the prompt engineering template of the language model; pushing the reconstructed task flow sequence to the task presentation terminal, controlling the task presentation terminal to display according to the updated rhythm, and locking the current presentation rhythm after detecting that the gaze disengagement rate time series has stabilized.

[0016] The steps of acquiring developmental norm data corresponding to the multimodal physiological signals and calculating the deviation feature of the source spatial features relative to the developmental norm data include: acquiring a set of physiological data of a healthy group of the same age in the same dimension; fitting the set of physiological data using a generalized additive model to generate a nonlinear age developmental trajectory curve; comparing the source spatial features with the corresponding reference points on the age developmental trajectory curve to generate standardized deviation values ​​as the deviation feature.

[0017] The method further includes: performing reverse mapping based on the developmental norm data to estimate the physiological brain age corresponding to the current multimodal physiological signal; calculating the difference between the physiological brain age and the actual age information of the subject to generate a brain age difference feature; and incorporating the brain age difference feature into the multimodal developmental deviation feature vector.

[0018] The steps for generating a multimodal developmental deviation feature vector include: extracting the actual physiological age of the subject; substituting the actual physiological age into the age development trajectory curve to calculate the distribution center location parameter and distribution dispersion parameter of the healthy peer group; dividing the difference between the actual measured value of the source spatial feature and the distribution center location parameter by the distribution dispersion parameter to obtain the dimensionless standardized deviation value; obtaining the second standardized deviation value of the subject's eye movement fixation feature and the third standardized deviation value of the heart rate variability feature; concatenating the standardized deviation value, the second standardized deviation value, and the third standardized deviation value along a preset dimension, and inputting them into a preset feature dimensionality reduction mapping layer for cross-modal redundancy removal processing, and outputting the orthogonalized multimodal developmental deviation feature vector.

[0019] A second aspect of this application provides a multimodal neurophysiological signal processing system, comprising:

[0020] A data acquisition module is used to passively acquire multimodal physiological signals of subjects in a resting state and an induced task state; wherein the multimodal physiological signals include at least electroencephalogram (EEG) signals;

[0021] An adaptive scheduling module, connected to the data acquisition module, is used to acquire the subject's cooperation characteristics and signal quality characteristics in the induced task state, and dynamically adjust the induced task content and presentation rhythm provided to the subject based on the cooperation characteristics and signal quality characteristics using a preset language model;

[0022] The feature extraction module, connected to the adaptive scheduling module, is used to perform source localization inverse solution on the collected EEG signals based on the template skull model of the subject's corresponding age group, and extract the source spatial features of the corresponding brain region.

[0023] The feature fusion module, connected to the feature extraction module, is used to acquire developmental norm data corresponding to the multimodal physiological signals, calculate the deviation features of the source space features relative to the developmental norm data, and fuse them to generate a multimodal developmental deviation feature vector.

[0024] The evaluation output module, connected to the feature fusion module, is used to input the multimodal developmental deviation feature vector into a preset deviation evaluation model for processing, and output the corresponding state deviation score data and data analysis report.

[0025] A third aspect of this application provides a computer-readable storage medium having a computer program stored thereon; when the computer program is executed by a processor, it implements the multimodal neurophysiological signal processing method described in the first aspect above.

[0026] The technical solution provided in this application has the following beneficial effects: It introduces a template skull model with age-varying conductivity parameters for source localization in the EEG signal analysis process, reducing the systematic electromagnetic wave attenuation error caused by skull thickness at different developmental stages, and improving the consistency and spatial accuracy of objective assessment of brain spatial signals. Simultaneously, by calculating the deviation characteristics of developmental norms, it removes background fluctuation interference caused by natural adolescent neural development. Furthermore, by using a language model to drive and dynamically schedule evoked tasks, it achieves a closed-loop response of the system to the real-time state of the subjects, controlling motion artifacts from the data acquisition source and improving compliance. Throughout the process, it outputs continuous state deviation scores through non-invasive objective data analysis, providing a quantitative reference for physiological signal data analysis scenarios. Attached Figure Description

[0027] Figure 1 This is a logical structure diagram of the multimodal neurophysiological signal processing system provided in the embodiments of this application.

[0028] Figure 2 This is a flowchart of the multimodal neurophysiological signal processing method provided in the embodiments of this application.

[0029] Explanation of reference numerals in the attached figures:

[0030] 101 Data acquisition module, 102 Adaptive scheduling module, 103 Feature extraction module, 104 Feature fusion module, 105 Evaluation output module. Detailed Implementation

[0031] To make the objectives, technical solutions, and beneficial effects of this application clearer, the application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be clarified that the signal processing method and system described in this embodiment aim to calculate, reduce the dimensionality, and map the deviation of objectively collected neurophysiological data, outputting continuous state deviation score data reflecting the fluctuation of physiological parameters. This system and method only provide objective quantitative reference information as a data processing and computer-aided evaluation tool; their results and output are calculation results and data analysis references based on physiological signals, and do not directly form medical diagnostic conclusions or treatment plans.

[0032] Please refer to Figure 2 , Figure 2 This is a flowchart of a multimodal neurophysiological signal processing method provided in an embodiment of this application. The method in this embodiment operates on a specific hardware platform, which mainly includes a multimodal sensor device, a synchronous trigger controller, a processor, and a communication interface. Figure 2 As shown, the method specifically includes steps S201 to S205.

[0033] Step S201: Passively collect multimodal physiological signals from the subject in both resting and induced task states; wherein the multimodal physiological signals include at least electroencephalogram (EEG) signals.

[0034] To comprehensively characterize the neurophysiological features of the subjects, the multimodal physiological signal acquisition equipment includes a functional magnetic resonance imaging (fMRI) scanner, an electroencephalogram (EEG) signal acquisition amplifier, and autonomic nervous system acquisition devices (such as heart rate variability recorders and electrodermal sensors). During signal acquisition, the system synchronously broadcasts aligned pulses to each modal sensor via a hardware timestamp module or a trigger pulse generator. The age range of the subjects can be set to a preset age group. Due to the developmental characteristics of underage subjects, directly placing them in a prolonged testing environment can easily induce anxiety, thereby compromising the quality of resting-state data. Therefore, the system performs a short resting-state acquisition session, followed by a switch to an evoked task state. In the evoked task state, the system presents subjects with multimedia contextual materials stratified by age group and calibrated for emotional valence and arousal.

[0035] Step S202: In the induced task state, the subject's cooperation characteristics and signal quality characteristics are obtained, and the induced task content and presentation rhythm provided to the subject are dynamically adjusted according to the cooperation characteristics and signal quality characteristics using a preset language model.

[0036] To address potential attentional distraction during the task, this application introduces a closed-loop adaptive adjustment mechanism based on a language model. The gaze disengagement rate is acquired using an infrared eye-tracking device deployed around the interactive display terminal, while the head movement frequency is acquired using a posture sensor integrated inside the EEG cap. The gaze disengagement rate and head movement frequency are jointly encoded to form a cooperation feature sequence. An impedance detection circuit in the EEG acquisition circuit extracts the contact impedance data of the current electrodes in real time as a signal quality feature.

[0037] The pre-configured language model is a time-series prediction network based on a long short-term memory network or a self-attention mechanism architecture. The language model establishes a continuous state memory sliding window in local memory, containing historical task execution records, the current gaze disengagement rate time series, and the current head movement frequency. The language model calculates and extracts the state deterioration gradient value within this sliding window. When the state deterioration gradient value exceeds a pre-set fatigue threshold, the system triggers a dynamic intervention mechanism. This mechanism inputs the state deterioration gradient value into the language model's prompt engineering template for context assembly, generating a reconstructed task flow sequence. This reconstructed task flow sequence includes instructions to simplify task elements, instructions to extend stimulus presentation duration, and positive visual feedback elements. This sequence is immediately pushed to the task presentation terminal. Upon receiving the instructions, the terminal hardware immediately changes the rendering rhythm and locks the current presentation rhythm after detecting that the subject's gaze disengagement rate time series has stabilized. Employing a closed-loop adaptive adjustment mechanism, the language model's logical reasoning ability can adapt to the subject's cooperation boundaries in real time, thereby suppressing electromyographic artifacts and signal loss caused by agitation at the source.

[0038] Step S203: For the collected EEG signals, source localization is reversed based on the template skull model of the subject's corresponding age group to extract the source spatial features of the corresponding brain regions.

[0039] In this step, to overcome the spatial aliasing problem caused by traditional techniques extracting features at the scalp electrode level, the system implements an electromagnetic field inverse solution based on a physical model. The subject's actual age information is obtained. A template skull model matching this actual age information is retrieved from a pre-built model database. The model database is pre-constructed based on structural magnetic resonance imaging data of a healthy group of the same age, mapping corresponding skull thickness and conductivity parameters for different age points. Considering that the skull is in a rapid ossification stage during adolescence, its water content is relatively high; therefore, the conductivity parameter in this age group differs significantly from the constant value in adults.

[0040] In the specific solution calculation, the template skull model is configured as a skull boundary element model. The system substitutes the skull thickness parameter and the age-varying conductivity parameter into this boundary element model to perform forward electromagnetic field modeling. The system calls a precise low-resolution EEG tomography algorithm to solve for the EEG signals. The specific physical calculation process of this algorithm is as follows:

[0041] A three-dimensional spatial coordinate system is established, incorporating the three-dimensional positional information of scalp electrodes and a skull boundary element model. A forward conduction lead field matrix, defined as variable K, is constructed based on conductivity parameters. This matrix reflects the electromagnetic attenuation mapping from the source space of the cerebral cortex to the scalp surface. A spatial smoothing prior assumption is introduced to calculate the electrode signal covariance matrix in the source space. Based on the forward conduction lead field matrix and the electrode signal covariance matrix, the system calculates the zero-error source localization weight matrix T. The system performs a dot product operation between the acquired EEG signal time series vector and the zero-error source localization weight matrix, ultimately estimating the source space current density. This can be achieved using the following formula:

[0042]

[0043] in, The output represents the three-dimensional dipole moment time series of the corresponding three-dimensional voxels in the source space, and its dimension is usually amperes per square meter; T represents the zero-error source localization weight matrix. This represents the continuous voltage time series vector of the acquired EEG signals. In a preferred embodiment, the source localization inverse solution process can use functional area activation signals as spatial prior constraints to improve the spatial accuracy of source spatial feature extraction. After obtaining the three-dimensional dipole moment time series, the system extracts the energy values ​​of specific frequency bands in the target brain region and uses them as source spatial features. Although this embodiment uses a precise low-resolution EEG tomography algorithm as an example, the source localization inverse problem solution algorithm is not limited to this. Those skilled in the art can also achieve the core objective of this application in extracting features in the source space using standard low-resolution EEG tomography algorithms, minimum norm estimation, or linearly constrained minimum variance beamforming algorithms. By using a template skull model of a specific age and the source localization algorithm, the dielectric constant that conforms to the physical characteristics of the subject's developmental stage can be used to eliminate the electromagnetic field attenuation error caused by differences in bone thickness, achieving high-precision source spatial signal reconstruction.

[0044] Step S204: Obtain developmental norm data corresponding to multimodal physiological signals, calculate the deviation features of source spatial features relative to developmental norm data, and fuse them to generate multimodal developmental deviation feature vectors.

[0045] Because the subjects are in a period of rapid neurodevelopment, the absolute values ​​of their physiological characteristics lack comparability across age groups. This embodiment introduces the calculation of deviations from developmental norm data. The system pre-acquires a set of physiological data of a healthy group of the same age under the same dimensions, and uses a generalized additive model to fit the physiological data set with a multi-parameter distribution, generating a nonlinear age development trajectory curve.

[0046] Regarding the calculation of deviation characteristics, the system extracts the actual physiological age value of the current subject, substitutes it into the age development trajectory curve, and calculates the distribution center location parameter of the healthy group of the same age. and the distribution dispersion parameter The system performs standardized deviation calculation, which involves subtracting the mean of the distribution corresponding to the age from the actual measured value of the source spatial characteristics and then dividing by the standard deviation of the distribution. The mathematical formula is as follows:

[0047]

[0048] Where z represents the dimensionless standardized deviation value of the calculated output, i.e. the deviation feature, which reflects the degree to which the feature deviates from the age group reference norm; x represents the specific measured value of the source space feature actually extracted in step S203. This represents the mean parameter of the distribution corresponding to the age group; This represents the standard deviation parameter of the distribution corresponding to age. By calculating the standardized deviation value z, the system removes baseline drift caused by natural aging in the feature values. For other modal signals such as eye movement fixation features, the low-frequency to high-frequency ratio in heart rate variability, and skin conductance response, the system simultaneously executes correction logic to obtain the second and third standardized deviation values, respectively. The system can also extract amygdala-to-prefrontal functional connectivity features, simultaneously execute the above correction logic, and obtain the corresponding standardized deviation values.

[0049] Based on this, the system performs nonlinear inverse mapping using full-scale developmental norm data to estimate the physiological brain age corresponding to the overall expression of current multimodal physiological signals. The system calculates the difference between this physiological brain age and the subject's actual physical age, generating a brain age difference feature. The system concatenates the first standardized deviation value, the second standardized deviation value, the third standardized deviation value, and the brain age difference feature along a preset one-dimensional tensor dimension, inputs them into a preset feature dimensionality reduction mapping layer for cross-modal redundancy removal, and outputs an orthogonalized multimodal developmental deviation feature vector. In an optional implementation, the preset feature dimensionality reduction mapping layer can also employ linear discriminant analysis or manifold learning algorithms. The application of the generalized additive model is only an example; quantile regression or multinomial regression algorithms can also be used to construct a dynamic baseline that changes with age.

[0050] Step S205: Input the multimodal developmental deviation feature vector into the preset deviation assessment model for processing, and output the corresponding state deviation score data and data analysis report.

[0051] The pre-defined deviation evaluation model is configured as a machine learning classifier deployed on the server backend. In this embodiment, the evaluation model employs an extreme gradient boosting tree model. During offline training, the model ingests labeled multimodal deviation data, and the objective function includes L1 and L2 regularization terms to suppress overfitting. After receiving the multimodal developmental deviation feature vectors, the evaluation model performs node splitting and weight accumulation through its internal tree structure, mapping continuous values ​​between 0 and 100 at the output layer as state deviation score data.

[0052] The system categorizes the degree of deviation based on preset age-group reference thresholds. It integrates state deviation score data, deviation level classifications, and deviation indices for key features of each modality to generate a structured data analysis report. For subjects with high deviation levels, the system prompts for a test-and-response consistency check within another time window. This assessment model uses feature norm deviation as input, a mechanism that allows the final output state deviation score to avoid the influence of scale differences in absolute values ​​across individuals of different ages, thus improving the statistical robustness of state deviation identification. This deviation assessment model can also be equivalently replaced by random forests, multimodal fusion networks, or graph neural networks.

[0053] This application provides a hardware architecture corresponding to the physical mapping of multimodal neurophysiological signal processing methods. Please refer to... Figure 1 , Figure 1 This is a logical structure diagram of the multimodal neurophysiological signal processing system provided in this application embodiment. The system includes: a data acquisition module 101, an adaptive scheduling module 102, a feature extraction module 103, a feature fusion module 104, and an evaluation output module 105.

[0054] The physical carriers of the data acquisition module 101 include a wearable EEG cap and an infrared eye tracker. The wearable EEG cap is used to acquire electroencephalogram (EEG) signals, and its internal impedance detection circuit is tightly embedded. The infrared eye tracker is positioned in front of the subject's line of sight and uses the principle of infrared light spot reflection to acquire the subject's eye movement trajectory. The above devices acquire the time series of multimodal physiological signals in real time through analog-to-digital conversion circuits. The data acquisition module 101 inputs the acquired signals into subsequent links.

[0055] The adaptive scheduling module 102 is electrically or communicatively connected to the data acquisition module 101. This module has a built-in microcontroller and display driver circuit, which is used to call a preset language model to process compliance features in the task initiation state, and modify the task content and presentation rhythm refreshed by the external display or virtual reality headset in real time.

[0056] The feature extraction module 103 is connected to the adaptive scheduling module 102. The core of the feature extraction module 103 is a digital signal processor or a graphics processor. This processor has a built-in matrix operation acceleration instruction set, which performs forward electromagnetic field modeling and brain power source localization inverse solving operations based on the template skull model, and solves the scalp electrical signals into cortical source spatial features.

[0057] The feature fusion module 104 is cascaded with the feature extraction module 103. The feature fusion module 104 relies on its internal large-capacity high-speed cache register to call developmental norm data pre-stored in non-volatile memory. It performs subtraction and division operations on the measured values ​​of source space features through its built-in arithmetic logic unit to extract the deviation features relative to the developmental norm data, and performs bus-level concatenation of deviation data from heart rate and eye movement to generate a multimodal developmental deviation feature vector.

[0058] The evaluation output module 105 is connected to the feature fusion module 104. This module is responsible for running an optimized lightweight deviation evaluation model and using a hardware multiply-accumulator network to output continuous state deviation score data from 0 to 100. The evaluation results are pushed to an external printing terminal via a network interface to generate a data analysis report.

[0059] To meet the demands of large-scale portable deployment, the hardware system incorporates an edge computing architecture. The system also includes edge computing nodes. These nodes connect to the wearable EEG cap and the infrared eye tracker via Bluetooth Low Energy or a proprietary radio frequency protocol. Each edge computing node has a built-in hardware acceleration unit dedicated to performing baseline drift removal processing of the EEG signals locally. After acquiring the raw voltage time series from the wearable EEG cap, the edge computing node constructs a circular data buffer in its local memory. Within this buffer, the node uses an adaptive filtering circuit to filter out 50Hz or 60Hz power frequency interference and EMG artifacts. After extracting the filtered net EEG data, the edge computing node uses a locally deployed hash encryption algorithm to de-identify and replace the header identifier of the net EEG data packet, generating an anonymized data packet. The edge computing node then sends this anonymized data packet to a cloud server via a wireless communication interface; the aforementioned feature extraction module 103, feature fusion module 104, and evaluation output module 105 are deployed on the backend cloud server. This edge computing isolation architecture reduces the power consumption and size requirements of front-end acquisition devices, while improving the isolation and data security of subject neural sensitive data processing at the physical connectivity level.

[0060] This application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is read and executed by a processor inside the system, it implements all the logical steps of the multimodal neurophysiological signal processing method. The storage medium can be constructed using solid-state storage or disk media.

[0061] The above description is merely a specific embodiment of this application, and the scope of protection of this application is not limited thereto. Those skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the above technical solutions. All such modifications and substitutions should be covered within the scope of protection of this application.

Claims

1. A multimodal neurophysiological signal processing method, characterized in that, include: Multimodal physiological signals of subjects were passively collected in both resting and induced task states; wherein the multimodal physiological signals included at least electroencephalogram (EEG) signals. In the induced task state, the cooperation characteristics and signal quality characteristics of the subject are obtained, and the induced task content and presentation rhythm provided to the subject are dynamically adjusted according to the cooperation characteristics and signal quality characteristics using a preset language model; The collected EEG signals are used to perform source localization inverse solution based on the template skull model of the subject's corresponding age group to extract the source spatial features of the corresponding brain regions. The developmental norm data corresponding to the multimodal physiological signals are obtained, the deviation features of the source spatial features relative to the developmental norm data are calculated, and the deviation features are fused to generate a multimodal developmental deviation feature vector. The multimodal developmental deviation feature vector is input into a preset deviation assessment model for processing, and the corresponding state deviation score data and data analysis report are output.

2. The multimodal neurophysiological signal processing method as described in claim 1, characterized in that, The steps for source localization and inverse solving based on the template skull model corresponding to the subject's age group include: Obtain the actual age information of the subject; The template skull model that matches the actual age information is retrieved from the preset model database; wherein the template skull model includes skull thickness parameters and electrical conductivity parameters corresponding to a specific age development stage.

3. The multimodal neurophysiological signal processing method as described in claim 2, characterized in that, The template skull model is configured as a skull boundary element model; the step of performing source localization inverse solution based on the template skull model corresponding to the subject's age group specifically includes: The skull thickness parameter and the electrical conductivity parameter are substituted into the skull boundary element model to perform positive electromagnetic field modeling; The EEG signal was solved using a precise low-resolution EEG to obtain the current density distribution of the scalp signal in the cortical source space.

4. The multimodal neurophysiological signal processing method as described in claim 3, characterized in that, The precise low-resolution EEG algorithm includes: Obtain a three-dimensional spatial coordinate system that includes scalp electrode location information and the skull boundary element model; A positively conducted lead field matrix is ​​constructed based on the conductivity parameters. The positively conducted lead field matrix reflects the electromagnetic attenuation mapping relationship from the source space of the cerebral cortex to the scalp surface. The covariance matrix of the electrode signals in the source space is calculated by introducing a spatial smoothing prior assumption; The zero-error source positioning weight matrix is ​​calculated based on the positively transmitted lead field matrix and the electrode signal covariance matrix; The continuous time series of the acquired EEG signal is multiplied with the zero-error source localization weight matrix to output the three-dimensional dipole moment time series of the corresponding three-dimensional voxels in the source space, and the frequency band energy of the three-dimensional dipole moment time series is used as the source space feature.

5. The multimodal neurophysiological signal processing method as described in claim 1, characterized in that, The steps for obtaining the subject's cooperation characteristics and signal quality characteristics include: The gaze disengagement rate of the subjects was obtained using an eye-tracking device; The subject's head movement frequency was obtained using a posture sensor; The gaze disengagement rate and the head movement frequency are combined to form the coordination characteristic.

6. The multimodal neurophysiological signal processing method as described in claim 5, characterized in that, The step of dynamically adjusting the content and presentation rhythm of the evoked task provided to the subject based on the cooperation characteristics and signal quality characteristics using a preset language model includes: The language model is used to perform time-series prediction of the gaze disengagement rate and the head movement frequency to generate fatigue trend indicators. Based on the fatigue trend index, a control command is issued to control the task presentation terminal to reduce the visual complexity of the induced task content and extend the interval of the presentation rhythm.

7. The multimodal neurophysiological signal processing method as described in claim 6, characterized in that, The step of dynamically adjusting the content and presentation rhythm of the evoked task provided to the subject based on the cooperation characteristics and signal quality characteristics using a preset language model further includes: Establish a continuous state memory sliding window that includes historical task execution records, the current gaze disengagement rate time series, and the current head movement frequency; Extract the state deterioration gradient values ​​within the continuous state memory sliding window; When the state deterioration gradient value exceeds the preset fatigue critical threshold, a dynamic intervention mechanism is triggered; The dynamic intervention mechanism generates a reconstructed task flow sequence containing task element simplification instructions, stimulus presentation duration extension instructions, and positive visual feedback elements by inputting the state deterioration gradient value into the prompt engineering template of the language model. The reconstructed task flow sequence is pushed to the task presentation terminal, which is then controlled to display the task at the updated pace. Once the time series of the gaze disengagement rate is detected to have stabilized, the current presentation pace is locked.

8. The multimodal neurophysiological signal processing method as described in claim 1, characterized in that, The step of acquiring developmental norm data corresponding to the multimodal physiological signals and calculating the deviation of the source spatial features from the developmental norm data includes: Obtain a set of physiological data on healthy individuals of the same age under the same dimensions; The physiological data set is fitted using a generalized additive model to generate a nonlinear age development trajectory curve; The source spatial features are compared with the corresponding reference points on the age development trajectory curve to generate standardized deviation values ​​as the deviation features.

9. The multimodal neurophysiological signal processing method as described in claim 1, characterized in that, The method further includes: Based on the developmental norm data, a reverse mapping is performed to estimate the physiological brain age corresponding to the current multimodal physiological signals; Calculate the difference between the physiological brain age and the subject's actual age information to generate brain age difference features; The brain age difference feature is incorporated into the multimodal developmental deviation feature vector.

10. The multimodal neurophysiological signal processing method as described in claim 8, characterized in that, The steps for fusing and generating multimodal developmental deviation feature vectors include: Extract the actual physiological age value of the subject currently being tested; Substitute the actual physiological age value into the age development trajectory curve to calculate the distribution center location parameter and distribution dispersion parameter of the healthy group of the same age; The difference between the actual measured value of the source spatial feature and the distribution center location parameter is divided by the distribution dispersion parameter to obtain the dimensionless standardized deviation value. The second standardized deviation value of the subject's eye movement fixation characteristics and the third standardized deviation value of the subject's heart rate variability characteristics were obtained; The standardized deviation values, the second standardized deviation values, and the third standardized deviation values ​​are concatenated along the feature dimension to form a one-dimensional feature vector, which is then input into a preset feature dimensionality reduction mapping layer for cross-modal redundancy removal processing, and the orthogonalized multimodal developmental deviation feature vector is output.

11. A multimodal neurophysiological signal processing system, characterized in that, include: A data acquisition module is used to passively acquire multimodal physiological signals of subjects in a resting state and an induced task state; wherein the multimodal physiological signals include at least electroencephalogram (EEG) signals; An adaptive scheduling module, connected to the data acquisition module, is used to acquire the subject's cooperation characteristics and signal quality characteristics in the induced task state, and dynamically adjust the induced task content and presentation rhythm provided to the subject based on the cooperation characteristics and signal quality characteristics using a preset language model; The feature extraction module, connected to the adaptive scheduling module, is used to perform source localization inverse solution on the collected EEG signals based on the template skull model of the subject's corresponding age group, and extract the source spatial features of the corresponding brain region. The feature fusion module, connected to the feature extraction module, is used to acquire developmental norm data corresponding to the multimodal physiological signals, calculate the deviation features of the source space features relative to the developmental norm data, and fuse them to generate a multimodal developmental deviation feature vector. The evaluation output module, connected to the feature fusion module, is used to input the multimodal developmental deviation feature vector into a preset deviation evaluation model for processing, and output the corresponding state deviation score data and data analysis report.

12. The multimodal neurophysiological signal processing system as described in claim 11, characterized in that, The data acquisition module includes: A wearable EEG cap is used to collect the EEG signals, and the wearable EEG cap has an embedded impedance detection circuit. An infrared eye tracker is used to acquire the eye movement trajectory of the subject.

13. The multimodal neurophysiological signal processing system as described in claim 12, characterized in that, The multimodal neurophysiological signal processing system also includes: Edge computing nodes are connected to the wearable EEG cap and the infrared eye tracker, respectively. The edge computing node has a built-in hardware acceleration unit for performing baseline drift removal processing of the EEG signal locally.

14. The multimodal neurophysiological signal processing system as described in claim 13, characterized in that, The edge computing nodes include: The raw voltage time series collected by the wearable brain electrode cap is obtained and a circular data buffer is constructed in the local memory space; An adaptive filtering circuit is used to filter out power frequency interference and electromyography artifacts within the circular data buffer. The filtered net EEG data is extracted, and the identity identifier of the net EEG data is desensitized and replaced by a locally deployed lightweight hash encryption algorithm to generate an anonymous data packet; The anonymized data packet is sent to the cloud server via a wireless communication interface; wherein the feature extraction module, the feature fusion module, and the evaluation output module are deployed on the cloud server.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the multimodal neurophysiological signal processing method as described in any one of claims 1 to 10.

Images