A real-time evaluation method and system for transcranial electrical stimulation effects

Abstract

The present invention discloses a real-time evaluation method for transcranial electrical stimulation effects, which relates to the technical field of brain electrical stimulation effects. By integrating MRI modeling, HD-tDCS electrode optimization, electric field simulation, and genetic algorithm parameter configuration, and combining multi-modal neural response acquisition, multi-task brain state prediction based on the attention mechanism, and Granger causal path modeling, the present invention constructs a real-time evaluation mechanism for the entire "stimulation → nerve → behavior" chain. Compared with the existing evaluation methods that only rely on electroencephalogram changes or single behavioral indicators, the present invention can simultaneously quantify the spatial targeting of stimulation, neural activation specificity, behavior improvement degree, and causal connectivity strength, and drive parameter self-feedback optimization through comprehensive scoring, significantly improving the real-time performance, accuracy, and individual regulation ability of transcranial electrical stimulation effect evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of electroencephalogram stimulation effect evaluation, and particularly to a real-time evaluation method and system for transcranial electrical stimulation effect. Background Art

[0002] With the rapid development of non-invasive neuromodulation technology, transcranial direct current stimulation has been widely applied to the intervention research of various neurological and psychiatric diseases such as depression, aphasia, chronic pain, and Parkinson's disease due to its high safety, simple operation, and potential for regulating brain function. In recent years, benefiting from the development of high-density electrode technology, the spatial resolution and targeting of stimulation have been improved. At the same time, combined with multimodal brain imaging and simulation modeling, the transformation from "group unified parameters" to "individualized stimulation configuration" has been realized. Currently, most HD-tDCS systems support three-dimensional electric field simulation on a structural MRI head model by means of electrode layout optimization, current intensity adjustment, etc., so as to achieve precise control of the stimulation area. In addition, physiological signals such as EEG, fNIRS, and EDA collected simultaneously during the stimulation of the subject also provide a basis for the analysis of the neural response pattern during the electrical stimulation process. However, despite the continuous optimization of stimulation parameters and the improvement of acquisition equipment, existing research generally uses offline analysis or one-dimensional indicators (such as power spectrum changes, reaction time improvement) for effect evaluation, and fails to achieve a dynamic evaluation closed-loop of "real-time, multi-dimensional, and feedbackable". Summary of the Invention

[0003] In view of the problems existing in the existing real-time evaluation method and system for transcranial electrical stimulation effect, the present invention is proposed.

[0004] Therefore, the present invention provides a real-time evaluation method and system for transcranial electrical stimulation effect to solve the problem that although the stimulation parameters are continuously optimized and the acquisition equipment is becoming more and more perfect, existing research generally uses offline analysis or one-dimensional indicators (such as power spectrum changes, reaction time improvement) for effect evaluation, and fails to achieve a dynamic evaluation closed-loop of "real-time, multi-dimensional, and feedbackable".

[0005] To solve the above technical problems, the present invention provides the following technical solutions:

[0006] In the first aspect, the present invention provides a real-time evaluation method for transcranial electrical stimulation effect, which includes,

[0007] Obtain the MRI data of the subject's head to construct a multi-tissue three-dimensional head model, deploy HD-tDCS needle-shaped composite electrodes on the head model, lock the target on the key functional areas of the brain, use the finite element method to simulate the electric field distribution, obtain the modulation maps of multiple sets of electrode current combinations, and use the genetic algorithm to optimize the electrode configuration and current parameters with the goal of maximizing the modulation intensity within the ROI and minimizing the diffusion in the non-target area;

[0008] Implement tDCS stimulation based on electrode configuration and parameters, collect test data in real time and perform preprocessing to construct the neural response vector R;

[0009] Construct a multi-task prediction network based on the attention mechanism to obtain the predicted value;

[0010] Guide the subject to perform a task feedback test, record the behavioral output B, and form a three-variable time-series data stream of the stimulation injection current intensity S, neural response R, and behavioral performance B. Adopt Granger causality analysis to quantify the causal path strength of S→R→B;

[0011] Calculate the EEG response enhancement rate and behavioral improvement rate, and calculate the comprehensive score of the stimulation effect based on the calculation results, the predicted value, and the causal path strength to perform real-time evaluation of the stimulation effect.

[0012] As a preferred solution of the real-time evaluation method for the transcranial electrical stimulation effect described in the present invention, wherein: acquire the subject's head MRI data to construct a multi-tissue three-dimensional head model, deploy HD-tDCS needle-shaped composite electrodes on the head model, target and lock the key functional areas of the brain, and use the finite element method to simulate the electric field distribution to obtain the modulation maps of multiple sets of electrode current combinations, including:

[0013] Acquire the subject's head T1-weighted structural image, output the DICOM format image sequence, convert the DICOM format to the NIfTI format, and unify the spatial resolution. Start the SPM12 software, select the "Segment" module to output the five-category tissue label map, and merge the channel label maps into a single label file;

[0014] Import the label map into ScanIP for voxelization reconstruction and tissue-level grid division to generate a complete three-dimensional finite element model of the head with conductive properties;

[0015] Activate the triangular patch grid in the scalp area as the electrode deployment surface, initialize the MNI space standard reference point, automatically project the international 10–10 system electrode points onto the scalp surface according to the head model space coordinates, adopt the normal projection algorithm, record the three-dimensional coordinates of each electrode, and assign a unique number to each electrode point to establish an electrode-coordinate index table, fixedly select TP9 as the reference electrode, and the remaining 60 electrode numbers form the candidate set of excitation electrodes;

[0016] At each electrode point Create a circular surface electrode model at the center, set the electrode material to silver, and define it as the "current injection boundary", set the reference electrode TP9 as the "current reference boundary", and fuse all electrode models with the head model surface grid to form a complete simulation geometric model;

[0017] Launch the COMSOL Multiphysics simulation platform and import the simulation geometric model;

[0018] Activate two independent excitation electrode pairs in the COMSOL model, set the current amplitude and alternating current with different injection frequencies for electrode pair 1 and electrode pair 2, fix the reference electrode TP9 as the common return electrode, and during the simulation, the two pairs of electrodes inject currents with different frequencies independently and simultaneously;

[0019] In the simulation software COMSOL, start the solver to calculate the electric field vector distributions formed by electrode pair 1 and electrode pair 2 in the brain tissue, denoted as 、 ;

[0020] The solver automatically traverses all the voxel grid points of the head phantom and outputs the numerical values of the two electric field vectors at each voxel coordinate, which are saved as three-dimensional field distribution data;

[0021] For all voxel positions, calculate the modulation amplitude at each voxel position based on the simulated electric field vectors, and generate a color visualization map of the three-dimensional spatial electric field intensity distribution.

[0022] As a preferred solution of the real-time evaluation method for the transcranial electrical stimulation effect described in the present invention, wherein: the genetic algorithm is used to optimize the electrode configuration and current parameters with the goal of maximizing the modulation intensity within the ROI and minimizing the non-target area diffusion, including:

[0023] Perform spatial registration on the subject's head phantom MRI image using the standard brain space, establish the mapping relationship between the physical coordinates and the standard template space in the simulation model, and map the spatial coordinates of the electric field distribution grid voxels into the template brain structure;

[0024] Extract the mask of the right hippocampal region from the standard brain template, match and extract all voxel indices located within the hippocampal mask according to the spatial registration coordinate table, record the grid numbers and spatial coordinates corresponding to the voxels, and form the ROI voxel set P;

[0025] Construct the ROI focusing ratio objective function D based on the electric field modulation amplitude data;

[0026] Randomly select 4 non-repeating numbers from the electrode candidate set, assign electrode numbers 1 to 4 in sequence, generate current injection values for electrode pair 1 and pair 2 respectively and sample according to a uniform distribution, combine the electrode numbers and current injection values into chromosomes, and store all the chromosomes as the initial population matrix after generation;

[0027] Configure electrode numbers and injection current parameters for each chromosome, solve the electric fields generated by the injection of two groups of electrodes respectively, calculate the amplitude of the interference modulation field, extract the electric field modulation value and voxel volume of each voxel point, calculate the total energy of the modulation field in the ROI region and the non-ROI region, and substitute it into formula D to obtain the fitness score of the current chromosome;

[0028] Construct a complete chromosome library with all individual results, sort them in descending order of fitness values, retain the first M chromosomes, and use the retained individuals as the next-generation population and define them as the elite subset Q, which does not participate in crossover and mutation;

[0029] Select non-elite individuals as crossover parents according to the fitness probability through roulette wheel selection and generate offspring chromosomes to obtain the newly generated offspring chromosome set A after crossover;

[0030] Randomly select W individuals from the offspring for mutation, and for each individual to be mutated, randomly select a strategy to adjust the parameters to obtain the mutated offspring set Z;

[0031] Merge and combine Q, A, and Z to construct a new generation of complete population, perform electric field simulation and fitness calculation on each chromosome in the new generation, stop evolution after reaching the preset maximum number of iterations, obtain the last generation of population, select the chromosome with the highest fitness in the last generation of population, extract the corresponding simulated electric field modulation map, and obtain the optimal electrode combination number and injection current parameters.

[0032] As a preferred solution of the real-time evaluation method for the transcranial electrical stimulation effect described in the present invention, wherein: guiding the subject to perform a task feedback test, recording the behavioral output B, and forming a three-variable time series data stream of the stimulation injection intensity S, the neural response R, and the behavioral performance B, and using Granger causality analysis to quantify the causal path strength of S→R→B includes:

[0033] Guide the subject to perform a Go / No-Go visual recognition task, record the reaction time and judgment correctness of the subject, and construct a time series behavior vector B(t);

[0034] After obtaining the B(t) behavioral feedback, immediately extract the stimulation injection intensity sequence S(t) and the neural response sequence R(t) within the corresponding time window;

[0035] Align the three types of data at the same sampling frequency and standardize them to form a three-variable time series Y(t), and use an a trous wavelet filter to decompose each component in the three-variable time series Y(t) separately to obtain the detail sequences of each variable at s scales;

[0036] Recombine the detail components of the three variables at each scale to obtain the joint sequence C(t) at each scale s, and extract from the joint sequence C(t) and , construct a two-layer Granger causal path model of "stimulus → nerve → behavior", including the first-stage modeling and the second-stage modeling;

[0037] Calculate the final path effectiveness score based on the results of the first-stage model and the second-stage model ;

[0038] Fuse the scores at all scales to obtain the final cross-scale main path score .

[0039] As a preferred solution of the real-time evaluation method for the transcranial electrical stimulation effect described in the present invention, wherein: the tDCS stimulation is implemented based on the electrode configuration and parameters, and the test data is collected in real time and preprocessed to construct the neural response vector R, including: configuring the stimulation system according to the determined optimal electrode pair combination and current injection parameters, collecting EEG, HbO / HbR dynamic concentration, EDA and pupil change data in real time and preprocessing them. After the preprocessing is completed, the processing results of each modality at each moment are combined into a structured neural response vector R.

[0040] As a preferred solution of the real-time evaluation method for the transcranial electrical stimulation effect described in the present invention, wherein: the construction of the multi-task prediction network based on the attention mechanism, and the obtained prediction value refers to the neural response vector sequence within the acquisition time window C , select the standard Encoder-only Transformer architecture, and input it into the trained Transformer model to obtain the cognitive response prediction value, the emotional fluctuation prediction value and the stimulation tolerance prediction value, and perform weighted summation on the obtained prediction values to obtain the neural state prediction score I.

[0041] As a preferred solution of the real-time evaluation method for the transcranial electrical stimulation effect described in the present invention, wherein: calculating the EEG response enhancement rate and the behavior improvement rate, and performing real-time evaluation of the stimulation effect based on the calculation results, the prediction value and the causal path strength refers to calculating the EEG response enhancement rate O and the behavior improvement rate F;

[0042] Perform weighted summation on the obtained EEG response enhancement rate O, behavior improvement rate F, and path strength score and the neural state prediction score I to obtain the real-time score J of the current stimulation effect;

[0043] Set the threshold K and , and K > , compare the real-time score J with the set threshold to obtain the current stimulation effect and formulate corresponding measures.

[0044] In a second aspect, the present invention provides a real-time evaluation system for transcranial electrical stimulation effects, including:

[0045] A three-dimensional modeling module for constructing an individual three-dimensional head tissue model based on MRI images and performing electrode positioning and mapping;

[0046] An electric field simulation module for performing electric field simulation based on the established head model and electrode layout and using a genetic algorithm to search for an optimal current injection configuration;

[0047] A signal acquisition module for implementing tDCS stimulation and simultaneously collecting multi-modal physiological signals;

[0048] A prediction module for constructing a Transformer network for the collected neural response sequence to achieve multi-dimensional brain state prediction;

[0049] A causal analysis module for guiding the subject to provide task feedback, extracting the stimulus-response-behavior three-way data stream, and performing Granger causal path modeling;

[0050] An effect evaluation module for fusing EEG response enhancement, behavior improvement, state prediction values, and causal path scores to calculate the current comprehensive stimulation effect score.

[0051] In a third aspect, the present invention provides a computer device including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the real-time evaluation method for transcranial electrical stimulation effects as described in the first aspect of the present invention is implemented.

[0052] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, and: when the computer program is executed by the processor, any step of the real-time evaluation method for transcranial electrical stimulation effects as described in the first aspect of the present invention is implemented.

[0053] The beneficial effects of the present invention are as follows: By integrating MRI modeling, HD-tDCS electrode optimization, electric field simulation, and genetic algorithm parameter configuration, combining multi-modal neural response acquisition, multi-task brain state prediction based on the attention mechanism, and Granger causal path modeling, a real-time evaluation mechanism for the entire "stimulus → nerve → behavior" chain is constructed. Compared with the existing evaluation methods that only rely on EEG changes or single behavioral indicators, the present invention can simultaneously quantify the spatial targeting of stimulation, neural activation specificity, behavior improvement degree, and causal connectivity strength, and drive parameter self-feedback optimization through comprehensive scoring, significantly improving the real-time performance, accuracy, and individual regulation ability of transcranial electrical stimulation effect evaluation. Description of the Drawings

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0055] Figure 1 It is a schematic flowchart of the real-time evaluation method for the transcranial electrical stimulation effect in Embodiment 1;

[0056] Figure 2 It is a schematic structural diagram of the real-time evaluation system for the transcranial electrical stimulation effect in Embodiment 1. Detailed implementation manners

[0057] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention with reference to the drawings in the specification.

[0058] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0059] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments.

[0060] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a real-time evaluation method for the transcranial electrical stimulation effect. The real-time evaluation method for the transcranial electrical stimulation effect includes the following steps:

[0061] S1. Obtain the head MRI data of the subject to construct a multi-tissue three-dimensional head model. Deploy HD-tDCS needle-shaped composite electrodes on the head model, target the key functional areas of the brain, use the finite element method to simulate the electric field distribution, obtain the modulation maps of multiple sets of electrode current combinations, and adopt the genetic algorithm to optimize the electrode configuration and current parameters with the goal of maximizing the modulation intensity within the ROI and minimizing the diffusion in the non-target area;

[0062] Specifically, obtaining the head MRI data of the subject to construct a multi-tissue three-dimensional head model, deploying HD-tDCS needle-shaped composite electrodes on the head model, targeting the key functional areas of the brain, and using the finite element method to simulate the electric field distribution to obtain the modulation maps of multiple sets of electrode current combinations includes:

[0063] Collect the T1-weighted structural images of the subject's head using a 3T magnetic resonance imaging system (such as Siemens Prisma), output the DICOM format image sequence, convert the DICOM format to NIfTI format using the MRIcron tool, and unify the spatial resolution. Start the SPM12 software, select the "Segment" module to output five types of tissue label maps, including gray matter (GM), white matter (WM), cerebrospinal fluid (CSF), bone (Skull), and Scalp (skin), and merge the channel label maps into a single label file. The label values are as follows: 1 = skin, 2 = skull, 3 = CSF, 4 = gray matter, 5 = white matter;

[0064] Import the label map into ScanIP (an advanced software platform developed by Simpleware, mainly used for processing, analyzing, and visualizing three-dimensional image data, especially data obtained from medical imaging technologies such as computed tomography (CT) and magnetic resonance imaging (MRI)), perform voxelization reconstruction and tissue-level mesh division to generate a complete three-dimensional finite element model of the head with conductive properties (such as gray matter 0.276 S / m, CSF 1.79 S / m, etc.);

[0065] Activate the triangular patch mesh in the scalp region as the electrode placement surface, initialize the five MNI space standard reference points: the tip of the nose (Nz), the occipital point (Iz), the left ear (A1), the right ear (A2), and the vertex (Cz). Complete the head model coordinate alignment through an automatic registration method (such as the three-point least squares method), project the international 10–10 system electrode points onto the scalp surface automatically according to the head model space coordinates, and use the normal projection algorithm to ensure the orthogonal landing points of the center point of each electrode on the scalp surface, record the three-dimensional coordinates of each electrode, and assign a unique number to each electrode point , the number range = 1~61, establish an electrode-coordinate index table;

[0066] Fixedly select TP9 (left mastoid) as the reference electrode, and the remaining 60 electrode numbers form a candidate set of excitation electrodes, which will be used to construct two sets of excitation electrode pairs in the subsequent combination optimization;

[0067] Use the modeling module to create a circular surface electrode model at the center of each electrode point , set the electrode material to silver and define it as the "current injection boundary", set the reference electrode TP9 to the "current reference boundary", and fuse all electrode models with the head model surface mesh to form a complete simulation geometry model;

[0068] Start the COMSOL Multiphysics simulation platform and import the simulation geometry model;

[0069] Activate two independent excitation electrode pairs within the COMSOL model, set the current amplitude and alternating current with different injection frequencies for electrode pair 1 and electrode pair 2, fix the reference electrode TP9 as the common return electrode. During the simulation, the two pairs of electrodes are independent and inject currents with different frequencies simultaneously to generate an interference modulation effect;

[0070] In the simulation software COMSOL, start the solver (Frequency Domain Solver) to calculate the electric field vector distributions formed by electrode pair 1 and electrode pair 2 in the brain tissue, denoted as 、 ;

[0071] The solver automatically traverses all the voxel grid points of the head phantom, outputs the numerical values of the two electric field vectors at each voxel coordinate, and saves them as three-dimensional field distribution data;

[0072] For all voxel positions, calculate the modulation amplitude at each voxel position based on the simulated electric field vectors:

[0073]

[0074]

[0075] In the formula, is the included angle between the two electric field vectors at any point in space, and are the electric field vectors generated by the two electrode pairs (electrode pair 1, pair 2) at the voxel point respectively, and are the magnitudes of the electric field vectors and respectively, is the maximum modulation envelope amplitude formed by the interference of the two alternating electric fields at the spatial position and represents the actual electric field interference stimulation intensity received at that position;

[0076] Use the three-dimensional field plotting module of the COMSOL simulation platform to directly map the modulation amplitude onto the three-dimensional head phantom grid to generate a complete three-dimensional spatial electric field intensity distribution color visualization map.

[0077] HD-tDCS refers to a technology that achieves high-spatial-resolution brain region regulation through a multi-electrode combination. It has better targeting compared to traditional large-area electrodes. The needle-shaped composite electrode introduces a multi-point micro-needle structure based on the traditional ring electrode or square electrode to form a current density focusing area, significantly enhancing the spatial resolution of the cortical surface electric field gradient distribution. The MNI space is a standard registration coordinate system for neuroimaging, used to ensure the unified alignment of head structures and electrode placement among different individuals. OMSOL Multiphysics is the most commonly used multi-physics field modeling platform in the current medical simulation field, supporting the solution of electromagnetic fields under different frequencies, boundaries, and medium properties. By introducing a 3T high-field-strength magnetic resonance device (such as Siemens Prisma) to obtain T1-weighted images and cooperating with the SPM12 and ScanIP toolchains, it is possible to achieve voxel-level mesh reconstruction of multiple head tissue structures (skin, bone, CSF, gray matter, white matter) based on medical images, providing anatomically realistic tissue boundary conditions and conductivity differences support for the simulation. Secondly, using the normal projection and electrode point-scalp mesh automatic alignment algorithm, on the premise of ensuring the MNI standard electrode system, the electrode layout is accurately mapped to the individual head model to form a complete electrode number - three-dimensional coordinate index library, providing a stable structural parameter basis for electrode optimization search. By introducing a dual-frequency alternating current injection mechanism in the COMSOL platform, setting the frequencies of electrode pair 1 and electrode pair 2 as f1 and f2 respectively, and controlling the phase and amplitude of the two groups of currents, the interference envelope in the deep cortical modulation region is maximized. Through the composite electrode structure + high-frequency cross-current injection mechanism + spatial vector modulation algorithm, the present invention realizes the construction of the physical optimal solution that maximizes the focusing of the three-dimensional modulation target area and minimizes the diffusion of the non-target area, laying a precise physical foundation for the subsequent formation of a closed-loop evaluation control system by combining neural signal modeling and behavioral feedback.

[0078] Further, using the genetic algorithm, with the goal of maximizing the modulation intensity within the ROI and minimizing the non-target area diffusion, optimize the electrode configuration and current parameters including:

[0079] Perform spatial registration on the subject's head model MRI image using the standard brain space (Talairach), establish the mapping relationship between the physical coordinates and the standard template space in the simulation model, and map the spatial coordinates of the electric field distribution grid voxels to the template brain structure;

[0080] Extract the mask of the right hippocampus region (Hippocampus_R) from the standard brain template (such as the Harvard-Oxford or AAL template of FSL), according to the spatial registration coordinate table, match and extract all voxel indices located within the hippocampal mask, record the grid numbers and spatial coordinates corresponding to these voxels, and form the ROI voxel set P;

[0081] Traverse all the voxel index sets B of the electric field simulation output, exclude the voxel numbers that already belong to P, and the remaining voxels form the non-target region set R. Also record their corresponding electric field values and voxel volumes;

[0082] Construct the ROI focusing ratio objective function D based on the electric field modulation amplitude data:

[0083]

[0084] In the formula, and are the three-dimensional spatial coordinates of the voxels in the ROI region (index u) and the voxels in the non-ROI region (index b), usually the coordinates of the center points of the voxels in the head model, is the set of voxel numbers of all voxels falling within the right hippocampal region (three-dimensional mask), representing the target region of electrical stimulation, is the set of voxel indices of all voxels that do not fall into the ROI mask region, representing the potential current diffusion region, is at each voxel position in the ROI region The maximum amplitude of the interference electric field modulation envelope formed after injecting current by the electrode pair, is at each voxel position in the non-ROI region The maximum amplitude of the interference electric field modulation envelope formed after injecting current by the electrode pair, and are the three-dimensional spatial volumes of the voxels. D is the ratio of the electric field modulation intensity in the ROI region to the diffusion intensity in the non-ROI region, and is used as the fitness function value of the genetic algorithm;

[0085] Randomly select 4 non-repeating numbers from the electrode candidate set, and assign electrode numbers 1 to 4 in sequence. Generate current injection values for electrode pairs 1 and 2 respectively and sample according to a uniform distribution. Among them, electrode numbers 1 and 2 form excitation electrode pair 1, and 3 and 4 form excitation electrode pair 2. Combine electrode numbers 1 to 4 and current injection values 1 to 2 into chromosomes, and store all chromosomes as the initial population matrix after generation;

[0086] Configure electrode numbers and current injection parameters for each chromosome, solve the electric fields generated by the two sets of electrode current injections respectively, calculate the interference modulation field amplitude, use the spatial registration table to correspond the simulation result voxels to the ROI voxel index set, extract the electric field modulation values and voxel volumes of each voxel point, calculate the total modulation field energy of the ROI region and the non-ROI region, and substitute them into the formula D to obtain the fitness score of the current chromosome;

[0087] Store the chromosome parameters and fitness together in a data structure, form a complete chromosome library with all individual results, sort them in descending order according to the fitness value, retain the first M chromosomes, and define the retained individuals as the next-generation population and the elite subset Q, which do not participate in crossover and mutation;

[0088] Select non-elite individuals as crossover parents according to the fitness probability through roulette wheel selection, set the current optimal individual. For each pair of parents, calculate the mean and variance of each parameter position by combining the current optimal individual. Generate offspring chromosomes using normal distribution sampling based on the obtained mean and variance. Round the electrode number parameter results in the offspring chromosomes to integers, and limit the ranges of the electrode number and current parameters to obtain the newly generated set A of offspring chromosomes after crossover;

[0089] Randomly select W individuals from the offspring for mutation. For each individual to be mutated, randomly select a strategy for parameter adjustment, including normal mutation, Cauchy mutation, and Lévy flight mutation. After mutation, re-limit the parameter ranges to obtain the mutated offspring set Z;

[0090] Combine Q, A, and Z to construct a new generation of complete population. Perform electric field simulation and fitness calculation on each chromosome in the new generation. Stop evolution after reaching the preset maximum number of iterations. Set the maximum number of iterations based on experimental optimization to obtain the last generation of population. Select the chromosome with the highest fitness in the last generation of population, extract the corresponding simulated electric field modulation map, and obtain the optimal electrode combination number and the injection current parameter (i.e., injection current intensity).

[0091] The "Talairach standard brain space" is an internationally recognized three-dimensional coordinate space system, commonly used in neuroimaging registration processing. Its advantage lies in the ability to uniformly map individual MRI image data to a standard template, making the results of electric field simulation have structural comparability and functional area consistency. The "mask" refers to a three-dimensional voxel set constructed for a specified functional area (such as Hippocampus_R) in the standard brain template, which is used to accurately extract the simulation results within this area and participate in the ROI target modeling. The "maximum amplitude of the electric field modulation envelope" refers to the peak value of the modulated electric field generated by the coherent superposition of two alternating currents with different frequencies in the body under the tTIS interference mechanism, representing the real stimulation intensity. By constructing the ROI focusing ratio objective function DDD, the present invention realizes the quantitative comparison of the total energy of electric field modulation in the target region (ROI) and the diffusion energy in the non-target region (non-ROI). Specifically, the larger this ratio, the more concentrated the stimulation energy is in the functional core area, and the smaller it is, the more significant the diffusion or deviation. This feature is set as the fitness function value of the genetic algorithm, giving the entire optimization process a clear physical objective-driven basis, thereby enhancing the scientific nature and interpretability of electrode selection and current injection strategies. In the optimization process, each group of "chromosome" coding represents a specific combination of electrode numbers and injected current values. Its initial population is constructed through number sampling and uniform current injection sampling, with coverage. The elite retention strategy ensures that local optimal solutions are not eliminated. The roulette wheel parent selection mechanism introduces diversity guidance with fitness probability control, and the crossover process uses parameter distribution fitting based on the current optimal body, so as to achieve iterative convergence of the fitness center without destroying the current structure solution space distribution. In addition, by adopting three mutation strategies of normal mutation, Cauchy mutation, and Lévy flight mutation, the synergy between local perturbation and global jump ability can be formed, improving the convergence speed of the algorithm and the ability to jump out of the local optimal trap.

[0092] S2. Implement tDCS stimulation based on the electrode configuration and parameters, and construct the neural response vector R after real-time collecting test data and preprocessing.

[0093] Specifically, tDCS stimulation is implemented based on electrode configuration and parameters. Test data is collected in real time and preprocessed to construct a neural response vector R. This includes configuring the stimulation system using a multi-channel HD-tDCS stimulation system (such as Starstim 64) according to the determined optimal electrode pair combination and current injection parameters. During the stimulation execution, an integrated physiological signal acquisition system (such as BIOPAC+NIRx) is used to collect EEG, dynamic HbO / HbR concentration, EDA, and pupil change data in real time. The EEG signal is filtered using a band-pass filter (0.5–70 Hz) to remove DC drift and EMG interference, and then independent component analysis (ICA) is used to remove eye movement and blink artifacts, effectively removing interference components such as eye movement and EMG, so that the remaining EEG data has a higher signal-to-noise ratio and interpretability validity. After the HbO / HbR signal is filtered by a sliding window and enhanced by differentiation, it can reflect the dynamic cerebral blood flow in the target area in real time, which helps to judge the regulatory effect of neuromodulation on the metabolic and blood oxygen coupling mechanisms. The data of the target channels are retained, and the low-frequency drift of the HbO / HbR dynamic concentration is removed based on the sliding window mean filter and then differentiated and enhanced. The EDA and pupil change data are first linearly interpolated to fill in the missing points and then normalized using Z-score to ensure their equal weight participation ability in the global modeling process. After the preprocessing is completed, the processing results of each modality at each moment are combined into a structured neural response vector R.

[0094] S3. Construct a multi-task prediction network based on the attention mechanism to obtain prediction values;

[0095] Specifically, constructing a multi-task prediction network based on the attention mechanism to obtain prediction values means using a sliding time window mechanism and setting the input window length to C. Therefore, each input contains the response vector of C time steps. , select the standard Encoder-only Transformer architecture, and the input is , the model includes an input embedding layer, a position encoding layer, a multi-head attention layer, a feed-forward network, and a multi-task output head. Define a loss function and an Adam optimizer to iteratively optimize the model parameters. When the preset maximum number of iterations is reached during continuous iterations, stop the iteration and output the updated model parameters of the Transformer model. After the model training is completed, the sequence of neural response vectors collected in real time is input into the trained Transformer model. The multi-head attention mechanism is used to process the time series features, and the three-way output branch is used to predict the three types of brain states of the subject, including the cognitive response prediction value, the emotional fluctuation prediction value, and the stimulation tolerance prediction value. The obtained prediction values are weighted and summed to obtain the neural state prediction score I.

[0096] By introducing a Transformer-based attention neural network structure during the real-time evaluation of tDCS, not only the parallel modeling and prediction of the multi-dimensional states of neural responses are realized, but also an interpretable and controllable scoring mechanism is constructed, providing an intelligent, dynamic, and quantifiable decision-making basis for the individualized regulation of the effects of electrical stimulation, with high innovation and engineering feasibility.

[0097] S4. Guide the subject to perform a task feedback test, record the behavioral output B, and form a three-variable time series data stream of the stimulus injection current intensity S, neural response R, and behavioral performance B. Use Granger causality analysis to quantify the causal path strength of S→R→B;

[0098] Specifically, guiding the subject to perform a task feedback test, recording the behavioral output B, forming a three-variable time series data stream of the stimulus injection current intensity S, neural response R, and behavioral performance B, and using Granger causality analysis to quantify the causal path strength of S→R→B includes:

[0099] Guide the subject to perform a Go / No-Go visual recognition task. Present an image every y seconds. The image types include target images and interference images. The subject needs to make a judgment within U seconds and respond by pressing a key. Record the reaction time and judgment correctness of the subject and construct a time series behavior vector B(t);

[0100] After obtaining the behavioral feedback of B(t), immediately extract the stimulus injection current intensity sequence S(t) and neural response sequence R(t) within the corresponding time window. S(t) and R(t) are constructed by extracting the corresponding stimulus injection current intensity and neural response combinations within the time window;

[0101] Align the three types of data at the same sampling frequency and standardize them to form a three-variable time series Y(t). Use an a trous wavelet filter to decompose each component in the three-variable time series Y(t) separately to obtain the detail sequences of each variable at s scales;

[0102] Recombine the detail components of the three variables at each scale to obtain a joint sequence C(t) at each scale s. Extract from the joint sequence C(t) and , and construct a two-layer Granger causal path model of "stimulus→nerve→behavior", including a first-stage modeling and a second-stage modeling;

[0103] The first-stage modeling is as follows:

[0104] Conduct Granger causality analysis of the stimulus on the neural response, construct two lag regression models, including a model with a stimulus term and a model without a stimulus term, and compare the residuals:

[0105]

[0106]

[0107] In the formula, and are the neural responses at scale s, p is the lag order of the neural response, automatically selected using the AIC criterion, is the regression coefficient (weight of the historical term), obtained by linear fitting using the least squares method, is the neural response value at the i-th past time step at scale s, is the model residual (error term), q is the lag order of the stimulus, is the regression coefficient of the stimulus lag term, solved by the least squares method, is the residual term of the model with the stimulus prediction, is the component of the electrical stimulation injection current intensity signal at the j-th lag time step at scale s;

[0108] According to the change of the residual variance, the Granger causal intensity of the stimulus on the neural response is:

[0109]

[0110] In the formula, and are the residual variances of models and ;

[0111] The second-stage modeling is as follows:

[0112] Perform Granger causal analysis of the neural response on the behavioral performance, and construct two behavioral regression models, including a model without neural response and a model with neural response:

[0113]

[0114]

[0115] In the formula, and are the detailed components of the behavioral response variable (such as RT) at scale s, and are the regression coefficients of the behavioral variable and the neural response variable, obtained through regression model training, and are the model residual terms, v is the lag order of the behavioral data, determined by the Akaike information criterion, p is the lag order of the neural response, is the wavelet detail component of the behavioral response variable (such as reaction time RT) at the i-th past time step under scale s, is the wavelet detail component of the neural response (such as EEG Δf band power) at the j-th lag point under scale s,

[0116] Calculate the Granger strength of the neural response to behavior as:

[0117]

[0118] Calculate the final path effectiveness score based on the results of the first-stage model and the second-stage model:

[0119]

[0120] where, is the effectiveness score of the closed path;

[0121] Fuse the scores at all scales to obtain the final cross-scale main path score :

[0122]

[0123] where S is the number of scale levels obtained by a trous wavelet decomposition, is the most significant path score at multiple scales.

[0124] Constructing a standardized three-variable time-series data stream through unified sampling frequency and standardization processing ensures that the three types of variables have an isochronous scale and a dimensionless analysis basis, which helps to improve the accuracy and stability of subsequent causal modeling. Using an a trous wavelet filter to perform multi-scale decomposition on each type of signal can capture potential stimulus-response characteristics layer by layer from low-frequency trends to high-frequency details. It not only retains the instantaneous changes caused by high-frequency stimuli but also retains the delayed regulatory effects within a long time window, providing a time-frequency dual-channel information structure for causal modeling. In the first-stage model, neural response regression models with and without stimulus terms are established respectively. By comparing the residual variances of the two models, the Granger causal strength of the stimulus on the nerve is defined, which can not only determine whether the stimulus effectively activates a specific neural response channel but also quantify its modulation intensity, thus forming neuro-effect evidence beyond the spatial electric field. In the second-stage model, a lag regression model of neural response to behavioral performance is constructed to determine whether there is an explanatory power of specific neural indicators (such as the Δf frequency band in EEG) on behavioral responses. This model can reveal whether the electrical stimulation truly "penetrates" the neural pathway and is transformed into an improvement at the behavioral level, distinguishing it from "ineffective behavioral" stimuli that only produce local neural activation but are ineffective in functional output. Finally, by fusing the causal strength scores of the two stages, a closed-path score S→R→B is constructed, and then the cross-scale main-path score is generated through the fusion of all scales, comprehensively measuring the continuity and closed-loop nature from stimulus emission, neural regulation to behavioral expression. The score result can not only be used for effect classification but also serve as a quantitative input signal for the feedback control mechanism to drive the adaptive adjustment of electrode layout or injection current parameters, realizing a truly closed-loop neural regulation system.

[0125] S5. Calculate the EEG response enhancement rate and the behavior improvement rate, and calculate the comprehensive score of the stimulation effect based on the calculation results, the predicted value, and the causal path strength to conduct real-time evaluation of the stimulation effect;

[0126] Specifically, calculating the EEG response enhancement rate and the behavior improvement rate, and calculating the comprehensive score of the stimulation effect based on the calculation results, the predicted value, and the causal path strength to conduct real-time evaluation of the stimulation effect means collecting the power spectral density o of the target frequency band under the current stimulation state and the average power spectral density of the same frequency band before the stimulation. Calculate the obtained EEG response enhancement rate O (subtract and divide by to get O), collect the current task execution reaction time and the average reaction time f before the task execution before the stimulation, and calculate the behavior improvement rate F (subtract and divide by f to get F);

[0127] The obtained EEG response enhancement rate O, behavior improvement rate F, and path strength score The weighted sum of the neural state prediction score I is used to obtain the real-time score J of the current stimulation effect;

[0128] Based on the ROC curve analysis, set the threshold K and , and K > , compare the real-time score J with the set threshold. If , then the current stimulation effect is significant, maintain the current stimulation parameters (electrode combination, current intensity, frequency unchanged), and continue the stimulation. If , then the stimulation effect is at a general level, automatically perform a fine-tuning operation, reduce the current injection intensity by 0.2 mA, and keep other parameters unchanged to prevent tolerance from decreasing, and enter the next cycle of evaluation. If , then the stimulation effect is poor, re-search for the optimal electrode pair combination and current injection scheme, replace the current configuration and enter the next round of simulation and stimulation.

[0129] The present invention particularly introduces two advanced indicators, the neural state prediction score (I) and the causal path strength score (G), to improve the comprehensiveness of the stimulation effect judgment and the explanatory power of the mechanism. The neural state prediction score encodes the current neural response vector sequence based on the Transformer attention network, outputs three types of predicted values: the cognitive state, emotional fluctuation, and stimulation tolerance of the subject, and forms the indicator I through weighted summation. This score can real-time perceive the changing trend of the subject's neural state and provide a predictive reference for "state perception - risk avoidance" during the electrical stimulation process. The causal path strength score G comes from the Granger causal analysis model, which models the causal dependence path among the three layers of variables "stimulation - neural response - behavioral output", reveals whether the behavior change caused by electrical stimulation is completed through a specific neural pathway, and quantifies the stimulation efficacy from the causal structure level. The system sets two thresholds K and , which are used to drive different levels of parameter control strategies. Through this score - judgment - feedback closed-loop mechanism, the present invention can dynamically adapt to the neural response characteristics of different individuals and achieve adaptive optimization control of the stimulation process.

[0130] This embodiment also provides a real-time evaluation method system for transcranial electrical stimulation effect, including:

[0131] A three-dimensional modeling module for constructing an individual three-dimensional head tissue model based on MRI images and performing electrode positioning and mapping;

[0132] An electric field simulation module for performing electric field simulation based on the established head model and electrode layout and searching for the optimal current injection configuration using a genetic algorithm;

[0133] A signal acquisition module for implementing tDCS stimulation and simultaneously collecting multi-modal physiological signals;

[0134] A prediction module, which is used to construct a Transformer network for the collected neural response sequence to achieve multi-dimensional brain state prediction;

[0135] A causal analysis module, which is used to guide the subject to give task feedback, extract the stimulus-response-behavior triple data stream, and perform Granger causal path modeling;

[0136] An effect evaluation module, which is used to fuse the EEG response enhancement, behavior improvement, state prediction value and causal path score to calculate the comprehensive effect score of the current stimulus.

[0137] This embodiment also provides a computer device, which is applicable to the situation of the real-time evaluation method of transcranial electrical stimulation effect, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the real-time evaluation method of transcranial electrical stimulation effect proposed in the above embodiment.

[0138] This computer device can be a terminal. This computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a carrier network, NFC (Near Field Communication) or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of this computer device can be a touch layer covered on the display screen, or a button, a trackball or a touchpad set on the computer device shell, or an external keyboard, touchpad or mouse, etc.

[0139] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for real-time evaluation of transcranial electrical stimulation effects proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc.

[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A real-time evaluation method for transcranial electrical stimulation effect, characterized in that: include, Obtain the MRI data of the subject's head to construct a multi-tissue three-dimensional head model, and place HD-tDCS needle-shaped composite electrodes on the head model, targeting the key functional areas of the brain. Use the finite element method to simulate the electric field distribution, and obtain the modulation spectrum of multiple sets of electrode current combinations. Use a genetic algorithm to optimize the electrode configuration and current parameters with the goal of maximizing the modulation intensity within the ROI and minimizing the diffusion of non-target areas. Implement tDCS stimulation based on electrode configuration and parameters, collect test data in real time and construct neural response vector R after preprocessing; Construct a multi-task prediction network based on the attention mechanism to obtain the predicted value, including the neural response vector sequence within the acquisition time window C , choose the standard Encoder-only Transformer architecture, Input the trained Transformer model to obtain the cognitive response prediction value, emotional fluctuation prediction value and stimulation tolerance prediction value, and perform weighted summation on the obtained prediction values ​​to obtain the neural state prediction score I; Guide the subjects to perform the task feedback test, record the behavioral output B, and form a three-variable time series data stream with the stimulus injection intensity S, neural response R, and behavioral performance B. Use Granger causal analysis to quantify the causal path strength of S→R→B; The EEG response enhancement rate and behavior improvement rate were calculated, and the comprehensive score of the stimulation effect was calculated based on the calculation results, the predicted value, and the causal path strength to conduct a real-time evaluation of the stimulation effect.

2. The real-time evaluation method of transcranial electrical stimulation effect according to claim 1, characterized in that: The method obtains the head MRI data of the subject to construct a multi-tissue three-dimensional head model, arranges HD-tDCS needle-shaped composite electrodes on the head model, targets key functional areas of the brain, and uses the finite element method to simulate the electric field distribution, and obtains modulation spectra of multiple sets of pole current combinations, including: Collect T1-weighted structural images of the subject's head, output DICOM format image sequences, convert DICOM format to NIfTI format, unify the spatial resolution, start SPM12 software, select the "Segment" module to output five types of tissue label maps, and merge the label maps of each channel into a single label file; The label map is imported into ScanIP for voxel reconstruction and tissue-level meshing to generate a complete three-dimensional finite element model of the head with conductive properties. Activate the triangular patch mesh of the scalp area as the electrode layout surface, initialize the MNI space standard reference point, automatically project the international 10-10 system electrode points onto the scalp surface according to the head model space coordinates, use the normal projection algorithm to record the three-dimensional coordinates of each electrode, and assign a unique number to each electrode point , establish an electrode-coordinate index table, select TP9 as the reference electrode, and the remaining 60 electrode numbers constitute the excitation electrode candidate set; At each electrode point Create a circular surface electrode model in the center, set the electrode material to silver, and define it as the "current injection boundary". Set the reference electrode TP9 as the "current reference boundary". Merge all electrode models with the head mold surface mesh to form a complete simulation geometry model. Start the COMSOL Multiphysics simulation platform and import the simulation geometry model; Two independent excitation electrode pairs are activated in the COMSOL model. The current amplitude and the AC current with different injection frequencies are set for electrode pair 1 and electrode pair 2. The reference electrode TP9 is fixed as the common return electrode. During simulation, the two pairs of electrodes are independently and simultaneously injected with currents of different frequencies. In the simulation software COMSOL, the solver is started to calculate the electric field vector distribution formed by electrode pair 1 and electrode pair 2 in the brain tissue, which are recorded as , ; The solver automatically traverses all the head model voxel grid points, outputs the values ​​of the two electric field vectors at each voxel coordinate, and saves them as three-dimensional field distribution data; For all voxel positions, the modulation amplitude of each voxel position is calculated based on the electric field vector obtained by simulation, and a color visualization of the three-dimensional electric field intensity distribution is generated.

3. The real-time evaluation method of transcranial electrical stimulation effect as claimed in claim 2, characterized in that: The genetic algorithm is used to maximize the modulation intensity within the ROI and minimize the diffusion in the non-target area, and the optimization of the electrode configuration and current parameters includes: The standard brain space is used to spatially register the MRI images of the subject's head model, and a mapping relationship between the physical coordinates and the standard template space is established in the simulation model, and the spatial coordinates of the electric field distribution grid voxels are mapped to the template brain structure; Extract the mask of the right hippocampus from the standard brain template, match and extract all voxel indexes within the hippocampus mask according to the spatial registration coordinate table, record the grid number and spatial coordinates corresponding to the voxel, and form the ROI voxel set P; The ROI focusing ratio objective function D is constructed based on the electric field modulation amplitude data; Randomly select 4 non-repeating numbers from the electrode candidate set, assign the electrode numbers 1 to 4 in sequence, generate current injection values ​​for electrode pair 1 and pair 2 respectively and sample them according to uniform distribution, merge the electrode numbers and current injection values ​​into chromosomes, generate all chromosomes and store them as the initial population matrix; Configure electrode numbers and injection parameters for each chromosome, solve the electric fields generated by the two sets of electrode injections, calculate the amplitude of the interference modulation field, extract the electric field modulation value and voxel volume of each voxel point, calculate the total amount of modulation field energy in the ROI area and non-ROI area, and substitute it into formula D to obtain the fitness score of the current chromosome; All individual results form a complete chromosome library, sort them in descending order of fitness value, retain the first M chromosomes, and define the retained individuals as the next generation population and the elite subset Q, which does not participate in crossover and mutation; Through roulette wheel selection, non-elite individuals are selected as crossover parents according to fitness probability and offspring chromosomes are generated to obtain the offspring chromosome set A generated after crossover; Randomly select W individuals from the offspring for mutation. For each individual to be mutated, randomly select a strategy to adjust the parameters to obtain the offspring set Z after mutation. Combine Q, A and Z to construct a new generation of complete population, perform electric field simulation and fitness calculation on each chromosome in the new generation, stop evolution after reaching the preset maximum number of iterations, and obtain the last generation of population. Select the chromosome with the highest fitness in the last generation of population, extract the corresponding simulated electric field modulation map, and obtain the optimal electrode combination number and injection current parameters.

4. The real-time evaluation method of transcranial electrical stimulation effect according to claim 3, characterized in that: The subject is guided to perform the task feedback test, the behavioral output B is recorded, the stimulus injection intensity S, the neural response R, and the behavioral performance B are formed into a three-variable time series data stream, and Granger causal analysis is used to quantify the causal path strength of S→R→B, including: Guide the subjects to perform the Go / No-Go visual recognition task, record the subjects' reaction time and judgment correctness, and construct the temporal behavior vector B(t); After obtaining the behavioral feedback B(t), the stimulus injection intensity sequence S(t) and the neural response sequence R(t) in the corresponding time window are immediately extracted; The three types of data are aligned and standardized at a unified sampling frequency to form a three-variable time series Y(t). The atrous wavelet filter is used to decompose each component of the three-variable time series Y(t) separately to obtain the detailed sequence of each variable at s scales. The detail components of the three variables at each scale are recombined to obtain the joint sequence C(t) at each scale s, and then extracted from the joint sequence C(t) and , construct a two-layer Granger causal path model of "stimulus→neuron→behavior", including first-stage modeling and second-stage modeling; Calculate the final path validity score based on the first-stage model and the second-stage model results ; The scores at all scales are integrated to obtain the final cross-scale main path score .

5. The real-time evaluation method of transcranial electrical stimulation effect according to claim 4, characterized in that: The method of implementing tDCS stimulation based on electrode configuration and parameters, collecting test data in real time and constructing a neural response vector R after preprocessing includes: configuring the stimulation system according to the determined optimal electrode pair combination and injection parameters, collecting EEG, HbO / HbR dynamic concentration, EDA and pupil change data in real time and preprocessing them, and after the preprocessing is completed, forming a structured neural response vector R with the processing results of each modality at each moment.

6. The real-time evaluation method of transcranial electrical stimulation effect according to claim 5, characterized in that: The calculation of the EEG response enhancement rate and the behavior improvement rate, the calculation of the comprehensive score of the stimulation effect based on the calculation results, the predicted value and the causal path strength, and the real-time evaluation of the stimulation effect refers to the calculation of the EEG response enhancement rate O and the behavior improvement rate F; The obtained EEG response enhancement rate O, behavior improvement rate F and path strength score The weighted sum of the predicted score of the neural state I is used to obtain the real-time score J of the current stimulation effect; Set the threshold K and , and K> , compare the real-time score J with the set threshold, obtain the current stimulation effect and formulate corresponding measures.

7. A real-time evaluation system for transcranial electrical stimulation effect, based on the real-time evaluation method for transcranial electrical stimulation effect according to any one of claims 1 to 6, characterized in that: include, A 3D modeling module, which is used to construct an individual 3D head tissue model based on MRI images and perform electrode positioning and mapping; The electric field simulation module is used to perform electric field simulation based on the built head mold and electrode layout and use genetic algorithm to search for the optimal injection configuration; A signal acquisition module, used to implement tDCS stimulation and simultaneously collect multimodal physiological signals; The prediction module is used to construct a Transformer network for the collected neural response sequences to achieve multi-dimensional brain state prediction; The causal analysis module is used to guide the subjects to perform task feedback, extract the stimulus-response-behavior triple data flow, and perform Granger causal path modeling; The effect evaluation module is used to integrate EEG response enhancement, behavior improvement, state prediction value and causal path score to calculate the comprehensive effect score of the current stimulation.

8. A computer device comprising: Memory and processor; The memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of the real-time evaluation method of the transcranial electrical stimulation effect according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for real-time evaluation of transcranial electrical stimulation effect according to any one of claims 1 to 6 are implemented.

Images