An electrical stimulation data processing method, system, device, and medium

Abstract

The application discloses an electric stimulation data processing method, system, device and medium. The method comprises the following steps: acquiring a baseline physiological signal of a subject, wherein the baseline physiological signal comprises multi-modal image data and resting-state electroencephalogram data, and extracting corresponding baseline features; controlling a stimulation device to output a detection electric stimulation signal to a target region, and controlling an acquisition device to synchronously acquire an electroencephalogram response signal; analyzing the electroencephalogram response signal to extract an envelope following response signal, and converting the envelope following response signal into a quantitative indicator indicating the participation degree of the target region; inputting the baseline features and the quantitative indicator into a response prediction model to output a prediction probability value of the detection electric stimulation signal; and generating a target control parameter according to the prediction probability value. The application can realize non-diagnostic and non-therapeutic system parameter optimization, provide pre-objective verification for the control of a neural regulation device, reduce parameter trial and error, and reduce costs.

Description

Technical Field

[0001] This application discloses a method, system, device, and medium for processing electrical stimulation data. The technical solution proposed in this application relates to the fields of physiological signal processing, device parameter control, and computer data processing. It performs in-depth data processing on baseline physiological signals and the real-time objective responses of subjects, outputting corresponding predicted response probabilities and target control parameters. It is primarily used for non-therapeutic regulation guidance and pre-device optimization verification. This solution belongs to a non-diagnostic data processing and device parameter recommendation system, and is particularly suitable for related data processing technology fields. Background Technology

[0002] Currently, the optimization of control parameters for neuromodulation devices mainly focuses on the design of transcranial magnetic stimulation (TMS) or deep brain stimulation (DBS). Most systems rely on baseline characteristics to directly infer parameters, lacking data feedback support specific to the unique physical mechanisms of temporal coherent electrical stimulation. This results in existing systems failing to objectively reflect the actual participation of modulation commands in deep intervention target areas, leading to poor applicability of the obtained stimulation parameters, significant individual differences in therapeutic efficacy, and high trial-and-error costs. Therefore, there is an urgent need for a data processing and control parameter optimization method based on real physical envelope-following response indicators for preliminary objective evaluation. Summary of the Invention

[0003] The main objective of this application is to provide an electrical stimulation data processing method, system, device, and medium to address the technical problem that existing neuromodulation systems rely excessively on static baseline prediction during parameter optimization, making it difficult to objectively reflect the participation of temporal coherent electrical stimulation in deep modulation, resulting in low parameter accuracy.

[0004] To achieve the above objectives, a first aspect of this application provides an electrical stimulation data processing method, comprising: acquiring a subject's baseline physiological signal and extracting baseline features corresponding to the baseline physiological signal; controlling a stimulation device to output a probe electrical stimulation signal to a target region and controlling a data acquisition device to simultaneously acquire the subject's electroencephalogram (EEG) response signal; analyzing the EEG response signal, extracting an envelope following response signal of a target frequency band, and converting the envelope following response signal into a quantitative index indicating the participation of the target region; inputting the baseline features and the quantitative index into a pre-trained response prediction model and outputting a predicted probability value for the probe electrical stimulation signal; and generating target control parameters for subsequent control of the stimulation device based on the predicted probability value.

[0005] Optionally, the extraction of baseline features corresponding to the baseline physiological signal includes: acquiring multimodal image data and resting-state EEG data; extracting structural connectivity features from the multimodal image data; extracting functional connectivity features from the resting-state EEG data; and concatenating the structural connectivity features and the functional connectivity features to obtain the baseline features.

[0006] Optionally, the control stimulation device outputs a probe electrical stimulation signal to the target area, including: controlling the stimulation device to output multiple high-frequency alternating current signals with frequency differences, wherein the multiple high-frequency alternating current signals spatially interfere in the target area to generate a time-domain coherent difference frequency envelope of the target frequency band; wherein the continuous output duration of the probe electrical stimulation signal is less than a first threshold.

[0007] Optionally, before analyzing the EEG response signal and extracting the envelope following response signal of the target frequency band, the method further includes: wherein the probe electrical stimulation signal includes a high-frequency carrier frequency band; filtering the EEG response signal using a preset artifact removal algorithm to filter out stimulation artifact signals corresponding to the high-frequency carrier frequency band, thereby obtaining a pure EEG signal; and the analysis of the EEG response signal includes analyzing the pure EEG signal.

[0008] Optionally, the step of filtering the EEG response signal using a preset artifact removal algorithm to remove stimulus artifact signals corresponding to the high-frequency carrier band and obtain a clean EEG signal includes: dividing the EEG response signal into equal-interval segments according to a first preset time step to obtain multiple signal segments; aligning all the signal segments on the time axis and calculating the mean amplitude to generate a local artifact template; subtracting the local artifact template from each of the signal segments to obtain a preliminary artifact-free signal; performing principal component analysis on the preliminary artifact-free signal to extract multiple independent principal components, identifying and removing the first principal component feature vector with a variance contribution rate greater than a second threshold and a frequency band distribution concentrated in the high-frequency carrier band; and performing inverse transformation reconstruction based on the retained principal component feature vectors to obtain the clean EEG signal.

[0009] Optionally, the step of analyzing the EEG response signal, extracting the envelope following response signal of the target frequency band, and converting the envelope following response signal into a quantitative indicator indicating the participation of the target region includes: performing bandpass filtering on the EEG response signal to retain the frequency band signal within the target frequency band; sequentially performing instantaneous amplitude envelope extraction and discrete Fourier transform on the frequency band signal to calculate the power spectral density at a specific difference frequency point within the target frequency band; and using the power spectral density at the specific difference frequency point as the quantitative indicator after numerical normalization.

[0010] Optionally, the calculation of the power spectral density at the specific difference frequency point and the conversion of the quantization index include: extracting the instantaneous amplitude envelope array of the frequency band signal using Hilbert transform; dividing the instantaneous amplitude envelope array into multiple time windows in the time domain, performing a Hanning window operation on the envelope data in each time window, and then performing a fast Fourier transform to obtain the amplitude spectrum tensor of each time window; calculating the element-wise sum of squares and average value of the amplitude spectrum tensor corresponding to each time window to obtain the average power spectrum; extracting the peak total energy of the average power spectrum within the specific difference frequency point and its adjacent preset frequency bandwidth limit; extracting the background noise frequency band energy corresponding to the no-stimulation period, and calculating the ratio of the peak total energy to the background noise frequency band energy; mapping the ratio to a continuous numerical range from zero to one, and outputting the mapping result as the quantization index.

[0011] Optionally, the step of inputting the baseline features and the quantization index into a pre-trained response prediction model and outputting a predicted probability value for the probed electrical stimulation signal includes: inputting the baseline features into a multilayer perceptron network for dimensionality reduction and compression to obtain a first hidden layer representation; inputting the quantization index into a linear mapping layer for processing to obtain a second hidden layer representation; calculating the feature cross-weighting coefficient matrix of the first hidden layer representation and the second hidden layer representation using an attention mechanism module; and inputting the fused feature vector multiplied by the feature cross-weighting coefficient matrix into a fully connected classifier to output the predicted probability value.

[0012] Optionally, before generating the target control parameters, the operation of obtaining the quantization index and the corresponding predicted probability value is performed cyclically under multiple pre-divided candidate space target point positions and multiple preset frequency differences; using the multiple quantization indices generated iteratively, a participation distribution spectrum is constructed with the candidate space target point positions and the preset frequency differences as index dimensions.

[0013] Optionally, generating target control parameters for subsequent control of the stimulation device based on the predicted probability value includes: selecting a set of candidate control combinations from the participation distribution spectrum whose predicted probability values ​​are greater than a third threshold; selecting a specific candidate spatial target location and a specific preset frequency difference from the set of candidate control combinations whose values ​​correspond to the largest quantitative index values; and sending the specific candidate spatial target location and the specific preset frequency difference as the target control parameters to the stimulation device.

[0014] Optionally, the electrical stimulation data processing method further includes: triggering an active verification process at preset time intervals during a continuous cycle of performing stimulation operations based on the target control parameters, controlling the stimulation device to output the probe electrical stimulation signal again, and acquiring the updated real-time EEG response signal.

[0015] Optionally, the electrical stimulation data processing method further includes: recalculating the current quantitative index based on the real-time EEG response signal, and sending the current quantitative index back to the response prediction model to dynamically update the current predicted probability value; when the current predicted probability value is lower than a preset safety value limit, triggering a parameter reset mechanism to regenerate and send closed-loop control parameters to the stimulation device.

[0016] A second aspect of this application provides an electrical stimulation data processing system, comprising: a multimodal baseline acquisition module configured to acquire baseline physiological signals of a subject and extract baseline features corresponding to the baseline physiological signals; a control synchronization module communicatively connected to the multimodal baseline acquisition module, configured to control a stimulation device to output probe electrical stimulation signals to a target region and control the acquisition device to synchronously acquire the subject's electroencephalogram (EEG) response signals; an engagement extraction module communicatively connected to the control synchronization module, configured to analyze the EEG response signals, extract envelope following response signals of a target frequency band, and convert the envelope following response signals into quantitative indicators indicating engagement in the target region; a response prediction module communicatively connected to the multimodal baseline acquisition module and the engagement extraction module, configured to input the baseline features and the quantitative indicators into a pre-trained response prediction model and output a predicted probability value for the probe electrical stimulation signals; and a parameter individualization module communicatively connected to the response prediction module, configured to generate target control parameters for subsequent control of the stimulation device based on the predicted probability value.

[0017] A third aspect of this application provides an electronic device, comprising: a memory for storing computer program instructions; and a processor electrically connected to the memory for executing the computer program instructions to implement the aforementioned electrostimulation data processing method.

[0018] A fourth aspect of this application provides a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the aforementioned electrical stimulation data processing method.

[0019] This application utilizes a short-duration temporal coherent probe electrical stimulation to simultaneously extract objective physical indicators of the envelope following response of the target region in a specific frequency band. These target region quantification features are then input into a model along with multimodal baseline features to construct a logic network for baseline state and dynamic physical feedback verification. Through this collaborative configuration, the excitation level of the interference target region can be objectively determined in advance, and control parameter combinations can be selected in reverse, reducing the time cost of trial-and-error iteration and improving the scientific rigor of parameter optimization and the control accuracy of the system. Attached Figure Description

[0020] Figure 1 This is a logical structure block diagram of the electrical stimulation data processing system provided in the embodiments of this application.

[0021] Figure 2 This is a flowchart of the electrical stimulation data processing method provided in the embodiments of this application.

[0022] Figure 3 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0023] To make the technical solutions and advantages of this application clearer, the specific embodiments of this application are described in detail below with reference to the accompanying drawings. It should be emphasized that the data processing and equipment control logic provided in this embodiment only outputs objective response probabilities and recommended operating parameters, which are intermediate process data used as a reference for technical personnel in decision-making. This system outputs recommended equipment parameters and objective data indicators, but does not directly form disease diagnosis conclusions or treatment plans. The data processing process and parameter distribution mechanism are located in the fields of equipment control and data processing.

[0024] Figure 2 This is a flowchart of an electrical stimulation data processing method provided in an embodiment of this application. The executing entity of this electrical stimulation data processing method can be a host computer device configured with a computing unit and a control bus, or a parameter recommendation engine server. (Reference) Figure 2 The electrical stimulation data processing method includes the following steps.

[0025] S201: Acquire the baseline physiological signals of the subject, wherein the baseline physiological signals include multimodal imaging data and resting-state EEG data, and extract the baseline features corresponding to the baseline physiological signals.

[0026] The multimodal baseline acquisition module 10 is responsible for acquiring multimodal neurophysiological signals of the subject in a static scene via a network interface or hardware acquisition card. This includes functional imaging data, high-density resting-state scalp EEG data, and peripheral electrocardiogram or skin conductance autonomic nerve signals. Upon receiving the raw signals, the system initiates a noise reduction and registration preprocessing procedure. For multimodal imaging data (e.g., diffusion tensor imaging data or functional magnetic resonance imaging data), the system calculates and extracts structural connectivity features of neural circuits related to the physiological state. The calculation of structural connectivity features is based on tracking the spatial distribution of white matter fiber tracts and quantitatively assessing the network connectivity strength of nodes. For resting-state EEG data, the system uses band-limited filtering and a phase-locked algorithm to extract functional connectivity features, reflecting the spontaneous electrophysiological synchronization properties between networks in different brain regions. After acquiring the two types of feature vectors, the microprocessor component 301 performs a data concatenation operation, splicing the structural connectivity features and the functional connectivity features in the matrix dimension to obtain a high-dimensional comprehensive baseline feature. By employing baseline features, a benchmark data profile can be established in terms of macroscopic anatomical structure and spontaneous network activity, providing a static input basis for the model. The method described above for obtaining comprehensive features through concatenation is an example; other equivalent algorithms such as autoencoders can also be used to achieve data integration.

[0027] S202: Control the stimulation device to output a probe electrical stimulation signal to the target area, and control the acquisition device to simultaneously acquire the subject's electroencephalogram (EEG) response signal.

[0028] After static feature acquisition is complete, the system enters active detection mode via the underlying control bus. The underlying system is equipped with a clock synchronization circuit, analog-to-digital and digital-to-analog converter arrays, and a communication bus interface based on a serial peripheral interface or Ethernet. The host computer main control program generates a digitized sequence of probe electrical stimulation waveform data and sends it to the control stimulation device, driving the device to output multiple high-frequency AC signals with frequency differences to the target test area.

[0029] Multiple high-frequency alternating current signals interfere spatially within deep neural tissue. Taking the interference of two sinusoidal electric fields as an example, the superposition relationship of their physical fields can be expressed by the following formula, and the system performs the calculation of the total interference electric field:

[0030]

[0031] Where E(t) represents the total interference electric field intensity received by the target region at time t; A1 and A2 represent the peak amplitudes of the electric fields of the first and second high-frequency carriers, respectively; and f1 represents the fundamental high-frequency carrier frequency. This is the characteristic frequency increment difference frequency parameter reflecting the frequency difference between the two. In this embodiment, The value range is 5Hz to 50Hz, used to generate the time-domain coherent difference frequency envelope of a specific target frequency band. Through spatial interferometry superposition, the high-frequency carrier forms a certain frequency in the deep intersection region. The low-frequency amplitude-modulated wave envelope. The capacitance integral characteristic of the neuronal cell membrane does not respond to high-frequency currents, but only induces charge accumulation at the point where the low-frequency time-domain coherent difference frequency envelope is received. During the detection process, the continuous output duration of the detection electrical stimulation signal is less than a preset first threshold, constituting an extremely short time-scaled quantitative detection.

[0032] While controlling the stimulation, the host computer sends a synchronous phase-locked acquisition command to the front-end EEG acquisition module. During the detection period, the synchronous EEG acquisition module 30 synchronously captures the subject's EEG response signal, including the artifact background, at a high-frequency sampling rate. By employing this synchronous control logic, it is ensured that the acquired EEG signal includes the physical changes induced by time-domain coherent electrical stimulation, avoiding data misalignment.

[0033] S203: Analyze the EEG response signal, extract the envelope following response signal of the target frequency band, and convert the envelope following response signal into a quantitative indicator indicating the participation of the target region.

[0034] The acquired raw EEG response signal contains high-amplitude, high-frequency electromagnetic interference caused by the stimulus source. The probe electrical stimulation signal includes a high-frequency carrier band. The response signal is filtered using a preset artifact removal algorithm to remove the stimulation artifact signal corresponding to the high-frequency carrier band and restore the pure EEG signal.

[0035] The artifact removal algorithm processing steps include: the system segments the original high-sampling-rate EEG response signal at equal intervals according to a first preset time step, obtaining multiple time-domain signal segments; the system aligns all signal segments on the time axis, extracts the envelope contour of periodic interference components by calculating the mean amplitude, and generates a local artifact template; the system subtracts the generated local artifact template from each signal segment to obtain the preliminary artifact-removed signal.

[0036] The system performs principal component analysis (PCA) for dimensionality reduction on the initially artifact-free signal. Microprocessor component 301 extracts multiple independent principal component eigenvectors and their corresponding eigenvalues. The logical relationship of the variance contribution rates of each principal component is as follows:

[0037]

[0038] in, This represents the variance contribution rate of the k-th independent principal component; This represents the extracted k-th feature value; n is the total decomposition dimension. The value is between 0 and 1. Based on a pre-set second threshold, the system inspects the extracted principal components, identifying and removing the first principal component eigenvectors whose variance contribution rate is greater than the second threshold and whose frequency band distribution is concentrated in the high-frequency carrier band. Based on the retained principal component eigenvectors, an inverse matrix transformation is performed to reconstruct the pure EEG signal. This algorithm can suppress measurement bias caused by electromagnetic interference.

[0039] After obtaining the clean EEG signal, the system uses a bandpass filter to filter and retain the frequency band signal within the target frequency range. By introducing the Hilbert transform algorithm, the instantaneous amplitude envelope array of the frequency band signal is extracted. The system divides the extracted instantaneous amplitude envelope array into multiple time windows in the time domain; after applying a Hanning window operation to the discrete envelope data sequence within each time window, a Fast Fourier Transform is performed to obtain the amplitude spectrum tensor for each time window; the sum of squares of the amplitude spectrum tensor elements corresponding to each time window is calculated and the average value is obtained to obtain the average power spectrum.

[0040] Based on the power spectrum distribution, the system extracts the specific difference frequency point. The peak total energy is measured within a very small, pre-defined frequency bandwidth limit (e.g., ±0.5 Hz). The background noise frequency band energy corresponding to the unstimulated period is extracted, and the ratio of the peak total energy to the background noise frequency band energy is calculated. This ratio is mapped to a continuous numerical range of [0,1], and the mapping result is output as a quantitative indicator. This indicator is the envelope follower response extraction identifier, which intuitively reflects the degree of bioelectrical participation in the target tissue.

[0041] S204: Input the baseline features and the quantification index into a pre-trained response prediction model, and output the predicted probability value of the probe electrical stimulation signal.

[0042] The response prediction module 50 performs system-level fusion of the acquired multimodal baseline features and target region participation quantification indicators, and inputs them into the pre-built response prediction evaluation model. The baseline feature vector data stream, which combines structure and function, is fed into a multilayer perceptron network for dimensionality reduction and compression to extract the first hidden layer representation; the quantification indicators are input into a linear mapping layer for processing to obtain the second hidden layer representation.

[0043] The pre-training process of the response prediction model includes the following steps:

[0044] Sample acquisition steps: Collect baseline characteristics and quantitative indicators of multiple sample subjects, and record the actual response status of the subjects under corresponding probe electrical stimulation, and use the actual response status as a label; wherein, the actual response status includes changes in motor threshold induced by electrical stimulation or subjective feeling rating, and the label is a binary marker;

[0045] Model building steps: Build an initial prediction model, which includes a multilayer perceptron dimensionality reduction sub-network, a linear mapping layer, an attention mechanism module, and a fully connected classifier. Initialize the weight parameters and bias parameters of each network layer.

[0046] Forward propagation steps: Input the sample baseline features into the multilayer perceptron dimensionality reduction sub-network to obtain the first sample hidden layer representation; input the sample quantization index into the linear mapping layer to obtain the second sample hidden layer representation; use the attention mechanism module to calculate the sample feature cross-weighting coefficient matrix of the first sample hidden layer representation and the second sample hidden layer representation; input the weighted fused sample fusion feature vector into the fully connected classifier and output the predicted response probability value.

[0047] Loss calculation steps: The difference between the predicted response probability value and the labeled value is calculated using the binary cross-entropy loss function;

[0048] Backpropagation steps: Based on the difference values, the weight parameters and bias parameters of each network layer of the initial prediction model are updated using the stochastic gradient descent optimization algorithm;

[0049] Iterative training steps: The forward propagation step, the loss calculation step, and the back propagation step are executed repeatedly until the difference value converges to less than the preset training termination threshold, thereby obtaining the response prediction model.

[0050] The model incorporates an attention mechanism module to calculate the feature cross-weighting coefficient matrix between the first and second hidden layer representations. The fused feature vectors are then fed into a fully connected classifier, which, along with a sigmoid activation function, outputs the predicted probability value for the probed electrical stimulation signal.

[0051] A conditional transfer module is configured within the control bus. The system acquires the predicted probability value and performs a judgment comparison. When the predicted result is lower than a specified threshold, the system forcibly blocks the parameter transmission process and provides an indication message at the front end. This prediction logic ensures effective regulation and safe control of subsequent equipment.

[0052] S205: Based on the predicted probability value, generate target control parameters for subsequent control of the stimulation device.

[0053] The parameter individualization module 60 generates target control parameters based on the predicted probabilities and the corresponding extracted quantization indicators. These target control parameters include the difference frequency parameter, the injected current intensity, and the candidate spatial target location. Before the system performs the parameter generation operation, iteratively acquires quantization indicators and corresponding predicted probability values ​​for multiple pre-divided candidate spatial target locations and multiple preset frequency differences. Using the iteratively generated quantization indicators, a participation distribution spectrum is constructed with the candidate spatial target location and the preset frequency difference as index dimensions.

[0054] The system filters out a set of candidate control combinations whose predicted probability values ​​are greater than a third threshold from the participation distribution spectrum. From this set, the system selects a specific candidate spatial target location and a specific preset frequency difference, along with the largest corresponding quantification index value. These locations and frequency differences are then sent to the stimulation device as target control parameters. This mechanism proactively locates the dedicated control parameters with the maximum response intensity, improving parameter optimization accuracy.

[0055] The system incorporates a closed-loop dynamic update and enhancement mechanism. During continuous cycles of stimulation based on target control parameters, an active verification process is triggered at preset time intervals. This process re-controls the stimulation device to output probe electrical stimulation signals and acquires updated real-time EEG response signals. The system recalculates the current quantification index based on the real-time EEG response signals and sends this index back to the response prediction model to dynamically update the current predicted probability value. When the current predicted probability value falls below a preset safety threshold, a parameter reset mechanism is triggered to regenerate and send closed-loop control parameters back to the stimulation device. This closed-loop mechanism endows the neuromodulation device with dynamic adaptive control capabilities.

[0056] Figure 1 This is a logical structure block diagram of the electrical stimulation data processing system provided in this application embodiment. This embodiment provides an electrical stimulation data processing system for carrying out and implementing the control tasks of the entire process described above.

[0057] The multimodal baseline acquisition module 10 is configured to read the sensor bus to acquire the baseline physiological signals of the subject and extract baseline features from them.

[0058] Control synchronization module V 01 It has a built-in high-frequency clock synchronizer and distributor, maintaining communication with the multimodal baseline acquisition module 10 and the underlying drive circuit. Control synchronization module V 01 The temporal coherence detection stimulation module 20 and the synchronous EEG acquisition module 30 are respectively coordinated and controlled. The synchronization module V is controlled. 01 The synchronous control time-domain coherent detection stimulation module 20 outputs a detection electrical stimulation signal; at the same time, it drives the analog-to-digital conversion pin of the synchronous EEG acquisition module 30 to acquire the EEG response signal.

[0059] The participation extraction module 40, whose high-speed processing unit is communicatively connected to the synchronous EEG acquisition module 30, is configured to perform filtering and analysis on the EEG response signal, extract the envelope following response signal component at the difference frequency, and convert it into a quantitative indicator of the participation in the target region.

[0060] The response prediction module 50 has its data input bus connected to the multimodal baseline acquisition module 10 and the participation extraction module 40. It is configured to load the spliced ​​baseline features and quantitative indicators into the forward classifier for calculation and output the predicted probability value.

[0061] The parameter individualization module 60 has its communication port connected to the output of the response prediction module 50. It is configured to analyze probabilistic scalars and, by integrating multi-point detection results, generates target control parameters.

[0062] The system provides a closed-loop update module 70, marked with a dashed box. The output V of the closed-loop update module 70 and the parameter individualization module 60... 02 The parameters are coupled and fed back to the front-end time-domain coherent detection stimulation module 20 and the synchronous EEG acquisition module 30 to achieve dynamic correction of the control parameters.

[0063] Figure 3 This is a schematic diagram of the hardware structure of the electronic device provided in this application embodiment. This embodiment provides an electronic device, the electronic device host V 03 The system includes a microprocessor component 301 (such as a central processing unit or graphics processing accelerator card), a physical memory module 302, and an input / output control interface card 303 that connects various transceiver units within the system via a communication bus. The microprocessor component 301 is electrically connected to the physical memory module 302 via circuitry. Upon startup, the microprocessor component 301 executes computer program instructions stored in sectors of the physical memory module 302, thereby implementing the aforementioned electrostimulation data processing method through the input / output control interface card 303.

[0064] This application provides a computer-readable storage medium. Computer program instructions are non-transitory stored above corresponding storage sectors of the computer-readable storage medium; when a processor executes this encoded sequence, it can implement the aforementioned electrical stimulation data processing method.

[0065] The embodiments described above are merely illustrative of the technical solutions of this application and are not intended to limit them. Those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. As long as the core data interaction logic of the inventive concept of this application is not departed from, it should be covered within the protection scope of this application.

Claims

1. A method for processing electrical stimulation data, characterized in that, include: The baseline physiological signals of the subjects were acquired, including multimodal imaging data and resting-state electroencephalogram data, and the baseline features corresponding to the baseline physiological signals were extracted. The stimulation device is controlled to output a probe electrical stimulation signal to the target area, and the acquisition device is controlled to simultaneously acquire the subject's electroencephalogram (EEG) response signal; wherein, the stimulation device is controlled to output multiple high-frequency alternating current signals with frequency differences, and the multiple high-frequency alternating current signals spatially interfere in the target area to generate a time-domain coherent difference frequency envelope of the target frequency band; The EEG response signal is analyzed to extract the envelope following response signal of the target frequency band, wherein the target frequency band is the time-domain coherent difference frequency band generated by the interference of multiple high-frequency AC signals, and the envelope following response signal is converted into a quantitative index indicating the participation degree of the target region. The baseline features and the quantification index are input into a pre-trained response prediction model, which outputs a predicted probability value for the probe electrical stimulation signal. Based on the predicted probability value, target control parameters are generated for subsequent control of the stimulation device.

2. The method as described in claim 1, characterized in that, The extraction of baseline features corresponding to the baseline physiological signal includes: Structural connectivity features are extracted from the multimodal image data; Extracting functional connectivity features from the resting-state EEG data; The baseline feature is obtained by splicing the structural connection feature with the functional connection feature.

3. The method as described in claim 1, characterized in that, Before analyzing the EEG response signal and extracting the envelope following response signal of the target frequency band, the method further includes: The detected electrical stimulation signal includes a high-frequency carrier band; The EEG response signal is filtered using a preset artifact removal algorithm to remove stimulation artifact signals corresponding to the high-frequency carrier band, thereby obtaining a clean EEG signal. The analysis of the EEG response signal includes the analysis of the pure EEG signal.

4. The method as described in claim 3, characterized in that, The step of filtering the EEG response signal using a preset artifact removal algorithm to remove stimulus artifact signals corresponding to the high-frequency carrier band and obtain a clean EEG signal includes: The EEG response signal is segmented at equal intervals according to a first preset time step to obtain multiple signal segments; All the aforementioned signal segments are time-axis aligned and the mean amplitude is calculated to generate local artifact templates; Subtract the local artifact template from each of the signal segments to obtain the preliminary artifact-free signal; Principal component analysis is performed on the preliminary artifact-removed signal to extract multiple independent principal components. The first principal component feature vector with a variance contribution rate greater than the second threshold and a frequency band distribution concentrated in the high-frequency carrier band is identified and removed. The pure EEG signal is obtained by inverse transformation reconstruction based on the retained principal component eigenvectors.

5. The method as described in claim 1, characterized in that, The step of analyzing the EEG response signal, extracting the envelope following response signal of the target frequency band, and converting the envelope following response signal into a quantitative indicator indicating the participation of the target region includes: The EEG response signal is bandpass filtered to retain the frequency band signal within the target frequency range; The instantaneous amplitude envelope is extracted and discrete Fourier transform is performed sequentially on the frequency band signal to calculate the power spectral density at a specific difference frequency point within the target frequency band. The power spectral density at the specific difference frequency point is obtained by numerical normalization and used as the quantization index.

6. The method as described in claim 5, characterized in that, The calculation of the power spectral density at the specific difference frequency point and the conversion process of the quantization index include: The instantaneous amplitude envelope array of the frequency band signal is extracted using Hilbert transform; The instantaneous amplitude envelope array is divided into multiple time windows in the time domain. After adding Hanning window operation to the envelope data in each time window, a fast Fourier transform is performed to obtain the amplitude spectrum tensor of each time window. Calculate the element-wise sum of squares and average value of the amplitude spectrum tensor corresponding to each time window to obtain the average power spectrum; Extract the total peak energy of the average power spectrum within the specific difference frequency point and its adjacent preset frequency bandwidth limit; Extract the background noise frequency band energy corresponding to the non-stimulated period, and calculate the ratio of the peak total energy to the background noise frequency band energy; The ratio is mapped to a continuous numerical range from zero to one, and the mapping result is output as the quantitative indicator.

7. The method as described in claim 1, characterized in that, The step of inputting the baseline features and the quantification index into a pre-trained response prediction model and outputting a predicted probability value for the probe electrical stimulation signal includes: The baseline features are input into a multilayer perceptron network for dimensionality reduction and compression to obtain the first hidden layer representation. The quantification index is input into the linear mapping layer for processing to obtain the second hidden layer representation; The feature cross-weighting coefficient matrix of the first hidden layer representation and the second hidden layer representation is calculated using the attention mechanism module; The fused feature vector, multiplied by the feature cross-weighting coefficient matrix, is input into the fully connected classifier, which outputs the predicted probability value.

8. The method as described in claim 1, characterized in that, Before generating the target control parameters, the operation of obtaining the quantization index and the corresponding predicted probability value is performed cyclically under multiple pre-divided candidate space target point positions and multiple preset frequency differences; Using multiple quantization indicators generated iteratively, a participation distribution spectrum is constructed with the difference between the candidate spatial target location and the preset frequency as the index dimension.

9. The method as described in claim 8, characterized in that, The step of generating target control parameters for subsequent control of the stimulation device based on the predicted probability value includes: From the participation distribution spectrum, a set of candidate control combinations whose predicted probability values ​​are greater than the third threshold is selected; In the set of candidate control combinations, select a specific set of candidate spatial target positions and a specific preset frequency difference corresponding to the largest value of the quantification index; The specific candidate spatial target location and the specific preset frequency difference are used as the target control parameters and sent to the stimulation device.

10. The method as described in claim 1, characterized in that, Also includes: During the continuous cycle of stimulation operation based on the target control parameters, an active verification process is triggered at preset time intervals to control the stimulation device to output the probe electrical stimulation signal again and acquire the updated real-time EEG response signal.

11. The method as described in claim 10, characterized in that, Also includes: The current quantitative index is recalculated based on the real-time EEG response signal, and the current quantitative index is sent back to the response prediction model to dynamically update the current predicted probability value. When the current predicted probability value is lower than the preset safety limit, the parameter reset mechanism is triggered to regenerate and send closed-loop control parameters to the stimulation device.

12. An electrical stimulation data processing system, characterized in that, include: A multimodal baseline acquisition module is configured to acquire the baseline physiological signals of the subject, wherein the baseline physiological signals include multimodal image data and resting-state electroencephalogram data, and to extract the baseline features corresponding to the baseline physiological signals; The control synchronization module is communicatively connected to the multimodal baseline acquisition module and is configured to control the stimulation device to output probe electrical stimulation signals to the target area and control the acquisition device to synchronously acquire the subject's electroencephalogram (EEG) response signals. The participation extraction module is communicatively connected to the control synchronization module and is configured to analyze the EEG response signal, extract the envelope following response signal of the target frequency band, and convert the envelope following response signal into a quantitative indicator indicating the participation of the target region. The response prediction module is communicatively connected to the multimodal baseline acquisition module and the participation extraction module, and is configured to input the baseline features and the quantification index into a pre-trained response prediction model, and output the predicted probability value of the probe electrical stimulation signal. The parameter individualization module is communicatively connected to the response prediction module and is configured to generate target control parameters for subsequent control of the stimulation device based on the predicted probability value.

13. An electronic device, characterized in that, include: Memory is used to store computer program instructions; The processor, electrically connected to the memory, executes the computer program instructions to implement the electrical stimulation data processing method as described in any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the electrical stimulation data processing method as described in any one of claims 1 to 11.

Images