High-precision transcranial direct current stimulation device based on navigation system

Abstract

The invention discloses a high-precision transcranial direct current stimulation device based on a navigation system, and relates to the technical field of electric digital data processing. In the method, an electroencephalogram data acquisition module in the high-precision transcranial direct-current stimulation device firstly acquires electroencephalogram data of a subject, a functional network analysis module immediately performs spectral analysis and functional connection analysis on the data, and the device compares the two analysis results with standard data of the same age group and then determines whether the subject is in the same age group. And identifying the abnormal function sub-network and quantifying the deviation degree. The navigation positioning module determines a multi-target stimulation position based on an individual three-dimensional navigation model and the identified abnormal sub-network, and the spatial accuracy of the multi-target transcranial direct current stimulation position is improved. The transcranial direct current stimulation scheme generation module takes the age, the deviation degree and the multi-target position as input characteristics, generates a personalized collaborative stimulation scheme through a machine learning model, and improves the accuracy of current parameter combinations corresponding to electrode layout schemes for different individuals in transcranial direct current stimulation.

Description

Technical Field

[0001] This application relates to the field of electrical digital data processing technology, and in particular to a high-precision transcranial direct current stimulation device based on a navigation system. Background Technology

[0002] Transcranial direct current stimulation (tDCS), as a non-invasive neuromodulation technique, modulates the excitability of neurons in the cerebral cortex by applying weak direct current to the scalp surface, and has broad application prospects in the fields of neurorehabilitation, cognitive enhancement, and treatment of mental illnesses.

[0003] Transcranial direct current stimulation (TCD) technology, a related field, primarily determines its stimulation protocol based on brain anatomy. It uses anatomical localization methods (such as the international 10-20 system or MRI guidance) to pinpoint the target brain region and places electrodes in the corresponding scalp area for stimulation. This method selects specific anatomical regions related to the target disease as stimulation targets based on known anatomical-functional correspondences of brain regions, achieving therapeutic goals by modulating neuronal excitability in these regions. Furthermore, the method optimizes and adjusts electrode placement and stimulation parameters based on feedback from the stimulation effect.

[0004] However, while anatomical structures provide information on the physical location of brain regions, they are insufficient to reflect the functional connectivity between different brain regions and abnormal patterns in the brain's functional network. When the target disease involves abnormal connectivity between multiple brain regions or disorder of the entire functional subnetwork, electrode layout schemes and corresponding current parameter combinations obtained based on a single or a few anatomical regions are not precise enough to effectively regulate the functional state of the entire abnormal network. Summary of the Invention

[0005] This application provides a high-precision transcranial direct current stimulation device based on a navigation system, which improves the accuracy of electrode layout scheme and corresponding current parameter combination.

[0006] Firstly, a high-precision transcranial direct current stimulation (TCD) device based on a navigation system is provided, comprising: an EEG data acquisition module for acquiring EEG data from a subject; a functional network analysis module for digitally filtering the EEG data and performing spectral analysis and functional connectivity analysis on the filtered EEG data, wherein the spectral analysis obtains the power proportion of each EEG wave at each electrode position, and the functional connectivity analysis constructs a functional connectivity network by calculating the functional connectivity strength between brain regions, and compares the power proportion and functional connectivity network with a preset standard EEG atlas and functional network pattern for the same age group to identify abnormal functional subnetworks and calculate the deviation of the power proportion and functional connectivity network; a navigation and positioning module for determining the location of multi-target TCD stimulation based on a three-dimensional navigation model and abnormal functional subnetwork constructed from the subject's individual brain structure imaging data; a TCD stimulation scheme generation module, including a preset machine learning model for generating a multi-target synergistic stimulation scheme using the subject's age, deviation, and multi-target TCD stimulation location as input features; and a TCD stimulation execution module for implementing the multi-target synergistic stimulation scheme on the subject.

[0007] By employing the above technical solution, the EEG data acquisition module first acquires the subject's EEG data. The functional network analysis module then performs spectral analysis and functional connectivity analysis on this data. Spectral analysis calculates the power proportion of each EEG wave at different electrode locations, while functional connectivity analysis constructs a functional network by calculating the connection strength between brain regions. Comparing these two analysis results with standard data for the same age group allows for the identification of abnormal functional subnetworks and the quantification of deviation. The navigation and positioning module determines multi-target stimulation locations based on the individual's 3D navigation model and the identified abnormal subnetworks, improving the spatial accuracy of multi-target transcranial direct current stimulation (TCD) locations. The TCD stimulation protocol generation module uses age, deviation, and multi-target locations as input features and generates personalized synergistic stimulation protocols through a machine learning model, improving the accuracy of current parameter combinations corresponding to electrode layout protocols for different individuals during TCD stimulation.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the training process of the preset machine learning model in the transcranial direct current stimulation (TCD) scheme generation module includes: a decision tree integration construction unit, used to construct a gradient boosting decision tree model based on the subject's age, deviation, multi-target TCD stimulation location, and effective historical multi-target synergistic stimulation schemes; a gradient boosting training unit, used to iteratively train using negative gradients as approximations of the residuals, training a new decision tree in each iteration to fit the residuals generated by the previous prediction; a feature importance evaluation unit, used to determine the weight contribution of the subject's age, deviation, and multi-target TCD stimulation location in the prediction by calculating the number of splits and information gain of each input feature in all decision trees; and a stimulation parameter output unit, used to weight and sum the prediction results of all decision trees according to a preset learning rate, and output a multi-target synergistic stimulation scheme composed of the current intensity, polarity, and stimulation duration parameters of each electrode.

[0009] By employing the above technical solutions, the gradient boosting training unit trains a new decision tree to fit the residuals of the previous iteration in each round, while the feature importance evaluation unit calculates the number of splits and information gain of each input feature in parallel. This allows the model training process to proceed synchronously with the feature weight determination process, reducing the total training time for model construction. The decision tree ensemble building unit combines multiple weak learners into a strong learner through the gradient boosting mechanism. Compared to a single prediction model, the ensemble model can capture the complex nonlinear relationship between age, deviation, and target location. The stimulation parameter output unit weights and sums all decision tree prediction results according to a preset learning rate, directly outputting precise values ​​for current intensity, polarity, and stimulation duration. This skips the multiple manual debugging and verification steps required in traditional methods, reduces uncertainty in the parameter determination process, and improves the prediction accuracy and generation efficiency of multi-target collaborative stimulation schemes.

[0010] In conjunction with some embodiments of the first aspect, some embodiments further include an individual difference adaptation module, comprising: a head geometric feature extraction unit, used to measure geometric parameters from the subject's brain structure imaging data, the geometric parameters including at least head circumference, skull thickness, cerebrospinal fluid thickness, and gray matter volume; an individualized conductivity calculation unit, used to calculate the conductivity of each layer of the subject's brain tissue based on the statistical relationship between age and brain tissue conductivity; an electric field conduction efficiency evaluation unit, used to calculate the electric field conduction efficiency coefficient from the electrodes to each target brain region using the finite element method based on the conductivity of each layer of the subject's brain tissue and head geometric features; and a stimulation parameter individualization adjustment unit, used to individualize the multi-target synergistic stimulation scheme output by the preset machine learning model according to the electric field conduction efficiency coefficient, increasing the current intensity of individuals with conduction efficiency less than a preset efficiency threshold by a preset adjustment ratio.

[0011] By employing the aforementioned technical solution, and through head geometric feature extraction and individualized conductivity calculation, a highly realistic head model in terms of anatomical morphology and physical properties was constructed for each subject. Based on this model, the electric field conduction efficiency assessment unit simulated the actual electric field distribution generated within the individual's head by the preset stimulation scheme using the finite element method, and quantified it as a conduction efficiency coefficient. Finally, the stimulation parameter individualization adjustment unit compensated for the current intensity based on this coefficient, thereby improving the accuracy of transcranial direct current stimulation current intensity for different individuals.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the electric field conduction efficiency evaluation unit includes: a finite element mesh generation sub-unit, used to divide the three-dimensional navigation model into tetrahedral mesh elements with side lengths no greater than a preset mesh size; a boundary condition setting sub-unit, used to map the electrode positions generated by the transcranial direct current stimulation scheme module onto the tetrahedral mesh elements constituting the scalp surface as voltage boundary conditions, and set other areas of the scalp as zero current density boundary conditions, the other areas being areas other than the electrode positions; a potential distribution solving sub-unit, used to solve the Laplace equation using the conjugate gradient method, and calculate the potential value of each mesh node by means of the constraint that the divergence between conductivity and potential gradient is zero; and a conduction efficiency calculation sub-unit, used to calculate the conduction efficiency coefficient by the ratio of the electric field gradient amplitude of the abnormal function sub-network to the electrode current.

[0013] By employing the above technical solution, the complex 3D navigation model is discretized into a high-quality tetrahedral mesh through finite element mesh generation sub-elements. Subsequently, the boundary condition setting sub-elements transform the physical stimulation (voltage or current) of the electrodes into mathematical constraints for the model. The potential distribution solution sub-elements use the efficient conjugate gradient method to solve the Laplace equation, improving the efficiency of calculating the potential distribution of the entire brain within a large mesh system.

[0014] In conjunction with some embodiments of the first aspect, some embodiments further include a data preprocessing module, comprising: an artifact identification unit, used to identify artifact components in the EEG data using an independent component analysis method, the artifact components including at least artifact components caused by blinking, electromyography, and power supply interference; an artifact removal unit, used to filter out artifact components according to a preset artifact removal threshold; a data quality assessment unit, used to score the quality of the preprocessed EEG data based on signal-to-noise ratio, spectral characteristics, and signal integrity; and a preprocessing parameter optimization unit, used to adjust the preset artifact removal threshold according to the data quality score results.

[0015] By employing the above technical solution, and utilizing independent component analysis (ICA) through an artifact identification and removal unit, noise interference in the EEG data was separated and eliminated. The data quality assessment unit quantifies the signal-to-noise ratio and integrity of the preprocessed data, forming an objective quality score. This score is then used as a feedback signal by the preprocessing parameter optimization unit to adaptively adjust the filtering and artifact removal parameters, thereby improving the purity of the EEG data.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, the functional network analysis module includes: an EEG signal extraction unit, used to extract EEG signals of preset frequency bands from EEG data, the preset frequency bands including at least delta waves, theta waves, alpha waves, beta waves, and gamma waves; a spectrum analysis unit, used to perform frequency domain transformation on the EEG data using fast Fourier transform, calculate the power spectral density of each electrode position in each frequency band, and normalize to obtain the power proportion of each EEG wave at each electrode position; a functional connectivity calculation unit, used to calculate the functional connectivity strength between brain regions in each frequency band using coherence analysis methods, and construct a multi-frequency functional connectivity network; a deviation calculation unit, used to statistically compare the power proportion and multi-frequency functional connectivity network of the subject with preset standard EEG atlases and preset functional network patterns of the same age group, and calculate the deviation; and an abnormal functional sub-network identification unit, used to identify combinations of brain regions with abnormal power distribution and abnormal connectivity patterns based on the deviation, and determine abnormal functional sub-networks.

[0017] By employing the aforementioned technical solution, the deviation calculation unit statistically compares the subject's power distribution and multi-band functional connectivity network with preset standard EEG atlases and preset functional network patterns for the same age group, converting physiological measurements into statistically significant deviations. Based on these deviations, the abnormal functional subnetwork identification unit improves the efficiency of integrating abnormal information from power distribution and connectivity patterns to determine abnormal functional subnetworks.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, the deviation calculation unit includes: a power percentage standard score calculation subunit, used to calculate the power percentage of each frequency band at each electrode location and the power percentage standard score of a preset standard EEG at the same age; a functional connectivity standard score calculation subunit, used to calculate the functional connectivity standard score of each frequency band functional connectivity strength between brain regions and the corresponding connection of a preset standard functional network pattern at the same age; and a comprehensive deviation matrix subunit, used to combine the power percentage standard score and the functional connectivity standard score to form a comprehensive deviation matrix, and use the comprehensive deviation matrix as the deviation.

[0019] By employing the aforementioned technical solution, the power percentage standard score calculation subunit and the functional connectivity standard score calculation subunit transform two different types and dimensions of brain functional indicators (power percentage and connectivity strength) onto a dimensionless common scale of standard scores through statistical methods. This allows for direct comparison and integration of the abnormality levels of the two indicators. The comprehensive deviation matrix subunit combines these two standard scores to form a structured comprehensive deviation matrix. This provides subsequent machine learning models with input features that have higher information density and richer dimensions, enabling the machine learning models to make decisions based on more comprehensive abnormal information, thereby improving the accuracy of stimulus protocol generation.

[0020] In conjunction with some embodiments of the first aspect, some embodiments further include a change calculation module, comprising: a post-stimulation data acquisition unit for acquiring EEG data of the subject at a preset time point after the stimulation ends; a corresponding analysis and processing unit for processing the post-stimulation EEG data through the same digital filtering, spectrum analysis, and functional connectivity analysis process as before stimulation; a band power change calculation unit for calculating the difference in power ratio before and after stimulation on an electrode-by-electrode and frequency-by-frequency-band basis, quantifying the degree of power change in each frequency band of each brain region; a network connectivity change calculation unit for calculating the strength change of corresponding connections in the functional connectivity network before and after stimulation, identifying connection patterns of enhanced, weakened, or restored normal connection strength; and a comprehensive change integration unit for integrating band power changes and network connectivity changes into a comprehensive neuromodulation effect evaluation index.

[0021] By employing the aforementioned technical solutions, the corresponding analysis and processing unit analyzes EEG data before and after stimulation under the same processing flow, reducing variables introduced by inconsistent processing methods. The band power and network connectivity change calculation unit quantifies the neurophysiological changes induced by stimulation at both the local and network levels. Finally, the comprehensive change integration unit merges these dispersed changes into a comprehensive evaluation index, providing a unified evaluation standard for comparing the effectiveness of different schemes.

[0022] In conjunction with some embodiments of the first aspect, some embodiments further include a feedback optimization module for updating the machine learning model using neural modulation effect evaluation indicators as feedback data.

[0023] By adopting the above technical solution, the evaluation index of neural modulation effect is used as feedback data to update the machine learning model, thereby improving the accuracy of the transcranial direct current stimulation scheme output by the transcranial direct current stimulation scheme generation module.

[0024] In conjunction with some embodiments of the first aspect, some embodiments further include a stimulation safety assurance module, comprising: a current density monitoring unit for monitoring the current density of the skin beneath each electrode; an impedance change detection unit for measuring the impedance of the electrode in contact with the skin at a preset detection frequency, and determining electrode detachment or poor contact when the impedance change value is greater than a preset change threshold; a temperature monitoring unit for monitoring skin temperature changes at the electrode site, and reducing the current intensity according to a preset adjustment ratio when the temperature is greater than a first preset temperature threshold; an emergency stop execution unit for cutting off the power supply to all electrodes within a preset response time when the detected current density is greater than a preset safe current density threshold, the impedance change value is greater than a preset change threshold, or the temperature is greater than a second preset temperature threshold; and a safety event recording unit for recording information on each safety event and establishing a safety file.

[0025] By employing the above technical solution, key physical parameters of the electrode-skin interface are monitored through three monitoring units: current density, impedance, and temperature. When the temperature exceeds a first threshold, the current is reduced proportionally to avoid unnecessary stimulation interruptions. When the detected current density exceeds a preset safe current density threshold, the impedance change exceeds a preset change threshold, or the temperature exceeds a second preset temperature threshold, the emergency stop unit directly cuts off all power, improving the safety of transcranial direct current stimulation.

[0026] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. The high-precision transcranial direct current stimulation (TCD) device first acquires the subject's EEG data through the EEG data acquisition module. The functional network analysis module then performs spectral analysis and functional connectivity analysis on this data. Spectral analysis calculates the power proportion of each EEG wave at different electrode locations, while functional connectivity analysis constructs a functional network by calculating the connection strength between brain regions. Comparing these two analyses with standard data for the same age group allows for the identification of abnormal functional subnetworks and the quantification of deviation. The navigation and positioning module determines the multi-target stimulation locations based on the individual's 3D navigation model and the identified abnormal subnetworks, thereby improving the spatial accuracy of multi-target TCD stimulation locations. The TCD stimulation protocol generation module uses age, deviation, and multi-target locations as input features and generates personalized synergistic stimulation protocols through a machine learning model, further improving the accuracy of current parameter combinations corresponding to electrode layout protocols for different individuals during TCD stimulation.

[0027] 2. Because the high-precision transcranial direct current stimulation device trains a new decision tree to fit the residuals of the previous iteration in each iteration through the gradient boosting training unit, while the feature importance assessment unit calculates the number of splits and information gain of each input feature in parallel, the model training process and the feature weight determination process are synchronized, reducing the total training time for model construction. The decision tree ensemble construction unit combines multiple weak learners into a strong learner through the gradient boosting mechanism. Compared with a single prediction model, the ensemble model can capture the complex nonlinear relationship between age, deviation, and target location. The stimulation parameter output unit weights and sums all decision tree prediction results according to a preset learning rate, directly outputting precise values ​​for current intensity, polarity, and stimulation duration. This skips the multiple manual debugging and verification steps required in traditional methods, thereby reducing the uncertainty in the parameter determination process and improving the prediction accuracy and generation efficiency of multi-target collaborative stimulation schemes.

[0028] 3. The high-precision transcranial direct current stimulation (TCD) device constructs a highly realistic head model for each subject in terms of anatomical morphology and physical properties through head geometric feature extraction and individualized conductivity calculation. Based on this model, the electric field conduction efficiency assessment unit uses the finite element method to simulate the actual electric field distribution generated within the individual's head by the preset stimulation scheme and quantifies it as a conduction efficiency coefficient. Finally, the stimulation parameter individualization adjustment unit adjusts the current intensity compensatorily based on this coefficient, thereby improving the accuracy of the TCD current intensity for different individuals. Attached Figure Description

[0029] Figure 1 This is a module architecture diagram of the high-precision transcranial direct current stimulation device in the embodiments of this application.

[0030] Figure 2 This is a schematic diagram of the data preprocessing module of the high-precision transcranial direct current stimulation device in the embodiments of this application.

[0031] Figure 3 This is a schematic diagram of the functional network analysis module of the high-precision transcranial direct current stimulation device in the embodiments of this application.

[0032] Figure 4 This is a schematic diagram of the deviation calculation unit of the high-precision transcranial direct current stimulation device in the embodiments of this application.

[0033] Figure 5 This is a schematic diagram of the individual difference adaptation module of the high-precision transcranial direct current stimulation device in the embodiments of this application.

[0034] Figure 6 This is a schematic diagram of the electric field conduction efficiency evaluation unit of the high-precision transcranial direct current stimulation device in this application embodiment.

[0035] Figure 7This is a schematic diagram of the stimulation safety protection module of the high-precision transcranial direct current stimulation device in the embodiments of this application.

[0036] Figure 8 This is a schematic diagram of the change calculation module of the high-precision transcranial direct current stimulation device in the embodiments of this application.

[0037] In the diagram, 101 is the EEG data acquisition module, 102 is the data preprocessing module, 103 is the functional network analysis module, 104 is the navigation and positioning module, 105 is the transcranial direct current stimulation (TCD) protocol generation module, 106 is the individual difference adaptation module, 107 is the TCD execution module, 108 is the stimulation safety assurance module, 109 is the change calculation module, 110 is the feedback optimization module, 1021 is the artifact recognition unit, 1022 is the artifact removal unit, 1023 is the data quality assessment unit, 1024 is the preprocessing parameter optimization unit, 1031 is the EEG signal extraction unit, 1032 is the spectrum analysis unit, 1033 is the functional connectivity calculation unit, 1034 is the deviation calculation unit, 1035 is the abnormal functional subnetwork identification unit, 10341 is the power percentage standard score calculation subunit, and 10342 is the functional connectivity standard score calculation subunit. The sub-units are as follows: 10343 is the comprehensive deviation matrix sub-unit; 1061 is the head geometric feature extraction unit; 1062 is the individualized conductivity calculation unit; 1063 is the electric field conduction efficiency evaluation unit; 1064 is the stimulus parameter individualized adjustment unit; 10631 is the finite element mesh generation sub-unit; 10632 is the boundary condition setting sub-unit; 10633 is the potential distribution solution sub-unit; 10634 is the conduction efficiency calculation sub-unit; 1081 is the current density monitoring unit; 1082 is the impedance change detection unit; 1083 is the temperature monitoring unit; 1084 is the emergency stop execution unit; 1085 is the safety event recording unit; 1091 is the post-stimulation data acquisition unit; 1092 is the corresponding analysis and processing unit; 1093 is the band power change calculation unit; 1094 is the network connection change calculation unit; and 1095 is the comprehensive change integration unit. Detailed Implementation

[0038] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0039] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0040] This application provides a high-precision transcranial direct current stimulation device based on a navigation system, which improves the accuracy of electrode layout scheme and corresponding current parameter combination.

[0041] Please see Figure 1 This is a module architecture diagram of the high-precision transcranial direct current stimulation device in the embodiments of this application.

[0042] This high-precision transcranial direct current stimulation device includes: The EEG data acquisition module 101 is used to acquire the EEG data of the subject.

[0043] Electroencephalography (EEG) data refers to the spontaneous, rhythmic electrical activity signals of neuronal populations in the cerebral cortex recorded by electrodes placed on the scalp; it is a direct reflection of brain function. The subjects indicated that they were individuals who underwent evaluation or regulation using this high-precision transcranial direct current stimulation device.

[0044] Specifically, during the data acquisition process, the operator first places a multi-channel EEG acquisition cap on the subject using an internationally standardized electrode positioning system (such as the 10-20 system), and injects conductive gel between the electrodes and the scalp to reduce the contact impedance to a preset low level (e.g., below 5 kiloohms). Then, the acquisition program is initiated, synchronously recording the voltage signals of all channels at a high sampling rate (e.g., 1000 Hz) and sufficient resolution (e.g., 24 bits) for a preset period of time (e.g., 5 minutes) to obtain sufficient data to reflect the subject's stable brain function during resting or task-oriented states.

[0045] The data preprocessing module 102 is used to preprocess the collected EEG data.

[0046] Preprocessing refers to a series of purification and correction operations performed on the raw EEG data, aiming to eliminate noise and artifacts of non-brain origin and convert it into a format suitable for subsequent analysis.

[0047] Specifically, firstly, the acquired continuous data is bandpass filtered, for example, by setting the passband to 0.5Hz to 45Hz, to filter out DC drift and high-frequency noise. Next, a 50Hz (or 60Hz) notch filter is applied to eliminate power line interference introduced by the power lines. Then, blind source separation techniques such as Independent Component Analysis (ICA) are used to decompose the mixed EEG signals into multiple independent components. By analyzing the waveform, spectrum, and scalp topography features of these components, artifacts representing eye movement (EOG), muscle activity (EMG), or electrocardiogram (ECG) are automatically or semi-automatically identified and removed or corrected from the raw data. Finally, the data is segmented, low-quality segments are removed, and all channels are rereferenced, for example, by converting to a whole-brain average reference, to eliminate bias caused by the selection of reference electrodes.

[0048] The functional network analysis module 103 performs spectral analysis and functional connectivity analysis on EEG data. The spectral analysis obtains the power proportion of each EEG wave at each electrode position. The functional connectivity analysis constructs a functional connectivity network by calculating the functional connectivity strength between brain regions. The power proportion and functional connectivity network are compared with preset standard EEG atlases and functional network patterns for the same age group to identify abnormal functional subnetworks and calculate the deviation of the power proportion and functional connectivity network.

[0049] Spectral analysis is used to represent the energy or power of different brain waves in different brain regions. Functional connectivity analysis quantifies the degree of synergy between signals from different brain regions by calculating the statistical correlation between them. Brain waves represent brain rhythms within a specific frequency range, such as delta waves, theta waves, and alpha waves, which are associated with different physiological and cognitive states. Power percentage refers to the relative proportion of the power of a specific brain wave in the total power. A functional connectivity network is a network model composed of brain regions (nodes) and the strength of functional connections (edges) between them. Preset standard EEG atlases and functional network patterns refer to normal power distribution maps and functional connectivity patterns obtained statistically from a large number of healthy age-matched individuals as a benchmark. Abnormal functional subnetworks represent the brain regions and their connections in the subject whose power or connectivity strength deviates from the normal standard. Deviation degree refers to a quantitative indicator of this deviation, such as a standard score.

[0050] Specifically, this module executes after data preprocessing. First, for each frequency band signal, the power spectral density at each electrode location is calculated using Fast Fourier Transform (FFT) or Welch's method, and then normalized to obtain the power proportion of each frequency band in each brain region. Second, indices such as coherence, phase synchronization index, or Granger causality between different electrode pairs in each frequency band are calculated to obtain a functional connectivity strength matrix, thereby constructing a multi-frequency band functional connectivity network. Then, the power proportion and functional connectivity strength of each frequency band in each brain region of the subject are statistically compared point-by-point and edge-by-edge with age-matched standard atlases and patterns stored in the database (e.g., calculating standard scores). This set of standard scores constitutes the deviation. Finally, brain regions and connections with deviations exceeding a preset threshold (e.g., |standard score|>1.96) are identified, forming a visualized abnormal functional subnetwork.

[0051] The navigation and positioning module 104 is used to determine the location of multi-target transcranial direct current stimulation based on the three-dimensional navigation model and abnormal functional subnetwork constructed from the individual brain structure imaging data of the subject.

[0052] Individual brain structural imaging data typically refers to high-resolution three-dimensional images obtained through magnetic resonance imaging (MRI) scans, capable of displaying the anatomical structures of the subject's brain, such as gray matter, white matter, and cerebrospinal fluid. A three-dimensional navigation model is a digital model reconstructed in a computer that matches the actual geometry of the subject's head and brain, serving as the basis for spatial localization. Multi-target transcranial direct current stimulation (DCS) locations refer to multiple specific coordinates determined on the scalp surface for placing stimulation electrodes, aiming to apply current at these locations to collectively modulate abnormal functional subnetworks.

[0053] Specifically, this module executes after the functional network analysis module identifies abnormal functional subnetworks. First, the subject's weighted MRI data is imported into the navigation software. Through image segmentation and surface reconstruction techniques, a multi-layered three-dimensional navigation model is generated, including the scalp, skull, and cerebral cortex. Next, the abnormal functional subnetworks identified in the previous step are projected or fused onto the individualized three-dimensional cerebral cortex surface using a spatial registration algorithm. Finally, considering the spatial distribution of the abnormal subnetworks, the placement of multiple anode and cathode electrodes that optimally cover these abnormal brain regions is planned automatically or under expert guidance on the scalp model, thus determining the multi-target stimulation locations.

[0054] The transcranial direct current stimulation protocol generation module 105 includes a preset machine learning model, which is used to generate a multi-target synergistic stimulation protocol using the subject's age, deviation degree, and multi-target transcranial direct current stimulation location as input features.

[0055] In this context, a pre-defined machine learning model refers to an algorithmic model trained on a large amount of historical input-output data, capable of simulating and predicting complex relationships, such as a gradient-boosting decision tree or a deep neural network. Input features represent the basis for the model's decision-making, including basic information about the subject, quantitative indicators of functional status, and stimulation target information. A multi-target synergistic stimulation protocol refers to a complete set of instructions containing specific parameters that need to be set on all target electrodes. These parameters typically include the current intensity, polarity (anode or cathode), and duration of stimulation for each electrode.

[0056] Specifically, the pre-set machine learning model uses historical subjects' age data, deviation data, and multi-target transcranial direct current stimulation location data, along with corresponding effective historical multi-target synergistic stimulation protocols, as a training set to learn the specific patterns of deviation data and electrode polarity and current intensity, the distribution of multi-target transcranial direct current stimulation locations, and the synergistic relationship of current between different electrodes. Age data serves as a moderating variable to fine-tune the intensity and duration of the overall protocol.

[0057] Based on historical subjects' deviation data and specific patterns of electrode polarity and current intensity, the distribution of multi-target transcranial direct current stimulation locations and the current synergy between different electrodes are analyzed. A pre-set machine learning model predicts a multi-target synergistic stimulation scheme based on the input subjects' age, deviation, and multi-target transcranial direct current stimulation locations.

[0058] The predicted multi-target co-stimulation scheme is then compared with the corresponding effective historical multi-target co-stimulation schemes, and the difference or error between the two is quantified by a pre-defined loss function. This loss function is minimized by continuously adjusting the parameters within the model (e.g., weights in the neural network or the structure of the decision tree).

[0059] By repeatedly performing this prediction-comparison-adjustment cycle on all training data, the pre-set machine learning model outputs a multi-target synergistic stimulation scheme that will continuously and systematically approximate historical multi-target synergistic stimulation schemes that have been verified as best practices.

[0060] The following is an example of a pre-trained Gradient Boosting Decision Tree (GBDT) regression model, and its implementation steps are as follows: 1. Input data vectorization processing: First, the module receives the deviation matrix (e.g., a 64×64 floating-point matrix) and flattens it, that is, converts it into a one-dimensional floating-point vector of length 4096 in row-major or column-major order.

[0061] Next, the module receives M three-dimensional position coordinates and concatenates these coordinates into a one-dimensional floating-point vector of length M×3 in a preset order (e.g., by electrode number).

[0062] Finally, the subject's age (a single value), the flattened deviation vector, and the concatenated position vector are concatenated into a one-dimensional feature vector X in the order determined during model training.

[0063] 2. Model loading and inference execution: Load a pre-trained and saved model file (e.g., a scikit-learn GBDT model object serialized via Python's joblib or pickle library) from local storage (such as a hard drive).

[0064] Calling the `predict` method of the model object, with the feature vector X generated during the input data vectorization process as the input parameter, i.e., executing `Y_raw = model.predict(X)`, allows the model to learn the potential relationships between different output parameters within a single model. Internally, the model will sequentially compute through hundreds of decision trees, and the outputs of all trees will be weighted and summed according to a preset learning rate, ultimately yielding a raw numerical output vector `Y_raw`. The dimension of this vector and the meaning of each element are predefined during the model training phase.

[0065] 3. Output vector parsing and post-processing: The module parses the original output vector Y_raw according to preset decoding rules. For example, assuming the model is designed for M electrodes, each electrode outputs two parameters (current intensity and polarity), and additionally outputs a global duration parameter, then the length of Y_raw is M×2+1.

[0066] This module processes the values ​​in Y_raw one by one: for current intensity values, boundary constraints are applied to trim them to a clinically safe range (e.g., [0.5, 2.0] mA), and quantization in 0.1 mA increments can be selected; for polarity values ​​(a continuous value), a threshold is applied, for example, values ​​greater than 0 are set as anodes, otherwise as cathodes. For duration values, range constraints (e.g., seconds) are also applied and rounded.

[0067] 4. Structured scheme generation: The parameters obtained after post-processing, each with a clear physical meaning, are combined with the corresponding electrode ID and location information to generate a structured data object (such as JSON format). This object is the final multi-target synergistic stimulation scheme, which is then sent to the transcranial direct current stimulation execution module (107) for execution.

[0068] The individual difference adaptation module 106 is used to adjust the multi-target synergistic stimulation protocol according to the individual physiological characteristics of the subject.

[0069] Here, individual physiological characteristics mainly refer to the individual-specific anatomical and biophysical parameters that affect the conduction of electric current in the head, such as head circumference, skull thickness, cerebrospinal fluid thickness, and the conductivity of different tissues.

[0070] Specifically, this module is executed after the transcranial direct current stimulation (TCD) protocol generation module generates the preliminary protocol but before stimulation is performed. First, using the individualized 3D head model constructed by the navigation and positioning module, key geometric parameters affecting the electric field distribution are measured. Simultaneously, based on the subject's age and other information, the conductivity of various brain tissue layers (such as the scalp, skull, and gray matter) is queried or estimated from a prior knowledge base. Then, based on these geometric and conductivity parameters, a rapid electric field simulation of the preliminary stimulation protocol is performed using the finite element method (FEM) to calculate the conduction efficiency coefficient of the current actually reaching the target brain region under the current parameters. Finally, the calculated conduction efficiency is compared with an ideal efficiency threshold. If it is lower than the threshold, the current intensity in the stimulation protocol is increased accordingly according to a preset ratio or functional relationship to compensate for conduction losses caused by individual differences.

[0071] The transcranial direct current stimulation execution module 107 is used to implement a multi-target synergistic stimulation protocol on the subject.

[0072] The implementation refers to the physical process of converting the digital stimulation scheme, which includes parameters such as current intensity, polarity, and duration, ultimately determined by the transcranial direct current stimulation scheme generation module 105 and the individual difference adaptation module 106, into a precise and stable analog direct current, and applying it to the subject through electrodes placed at the target point on the scalp.

[0073] Specifically, this module is executed after the transcranial direct current stimulation protocol generation module 105 and the individual difference adaptation module 106 are completed. First, the operator places and fixes the stimulation electrodes on the subject's scalp according to the location provided by the navigation and positioning module. Then, the finalized multi-target synergistic stimulation protocol is loaded into the stimulator's main control unit. When stimulation is initiated, multiple independent constant current sources inside the stimulator precisely output direct current of specified intensity and polarity to each electrode channel according to the parameters set in the protocol. Throughout the stimulation process, the main control unit strictly controls the total stimulation duration and automatically and smoothly stops the current output after the preset time is reached, thus completing one full stimulation cycle.

[0074] The Stimulation Safety Assurance Module 108 is used to monitor and ensure the safety of the subject during the stimulation process.

[0075] The change calculation module 109 is used to quantitatively evaluate the effect of neural regulation after the stimulation ends.

[0076] Specifically, this module is activated after the transcranial direct current stimulation (TCD) execution module completes a full stimulation cycle. Its workflow is as follows: First, at a preset time point after the stimulation ends (e.g., 5 minutes later), the EEG data acquisition module 101 acquires resting-state EEG data under conditions identical to those before stimulation. Then, this post-stimulation EEG data is fed into the same data preprocessing module 102 and functional network analysis module 103 processes as before stimulation to obtain the post-stimulation power ratio and functional connectivity network. Finally, the post-stimulation and pre-stimulation analysis results are compared pairwise: the difference in power ratio is calculated electrode-by-electrode and frequency band-by-frequency band to obtain the band power change; the difference in functional connectivity strength is calculated connection-by-connection to obtain the network connectivity change. These changes can be further integrated into one or more comprehensive neuromodulation effect evaluation indicators.

[0077] The feedback optimization module 110 is used to update the machine learning model with the evaluation index of neural modulation effect as feedback data.

[0078] Feedback data refers to a set of data records that contain complete information about a stimulus (including the state before the stimulus and the stimulus protocol used) and the exact effect produced by the stimulus (i.e., the evaluation index of neural modulation effect). Updating the machine learning model refers to the process of retraining or fine-tuning the pre-set machine learning model using new feedback data to continuously improve its predictive ability and decision-making level, which is the key to achieving system self-adaptation and intelligence.

[0079] Specifically, this module is triggered after the change calculation module produces a quantitative evaluation index. First, the complete data of this modulation is packaged into a new training sample. This sample includes: the subject's age as input features, the deviation before stimulation, the multi-target stimulation location and stimulation parameters used, and the neuromodulation effect evaluation index calculated by the change calculation module as the output label (ground truth). Then, this new training sample is added to the model's historical training dataset. Finally, according to a preset strategy (e.g., every 10 new data points collected, or during daily free time), this updated dataset can be used to incrementally train or fully retrain the machine learning model in the transcranial direct current stimulation protocol generation module 105. Through this process, the model continuously learns from new successful or unsuccessful cases, thus enabling it to generate more accurate and effective stimulation protocols when facing similar situations in the future.

[0080] The above describes the modular architecture of the high-precision transcranial direct current stimulation device in the embodiments of this application. The modules are described below.

[0081] Please refer to the following: Figure 2 This is a schematic diagram of the data preprocessing module of the high-precision transcranial direct current stimulation device in this application embodiment.

[0082] Data preprocessing module 102 includes: The artifact identification unit 1021 is used to identify artifact components in EEG data using an independent component analysis method. The artifact components include at least artifact components caused by blinking, electromyography, and power supply interference.

[0083] Independent Component Analysis (ICA) is a signal processing technique for blind source separation. Its core idea is to assume that the mixed signal recorded by scalp electrodes is a linear superposition of multiple statistically independent source signals (including brain neuronal activity signals and artifact signals from non-brain sources). The method aims to find the inverse process of this mixing, thereby separating these independent source signals. Artifact components are independent components separated by ICA whose time-domain waveforms, frequency-domain characteristics, or spatial topography conform to typical characteristics of non-brain activities such as blinking, muscle activity, ECG, or power line interference. Blink artifacts typically appear as high-amplitude, low-frequency pulse-like signals occurring in the forehead region. Electromyography (EMG) artifacts appear as high-frequency, spike-like noise, often appearing on electrodes near muscles (such as the temporal and frontal regions). Power line interference artifacts appear as sinusoidal waveforms with a peak value at 50 Hz or 60 Hz.

[0084] Specifically, this unit executes after preliminary bandpass filtering of the EEG data. Its workflow is as follows: First, the pre-filtered multi-channel EEG data matrix is ​​input into the ICA algorithm model (such as the commonly used Infomax or FastICA algorithm). The algorithm iteratively calculates to find an optimal unmixing matrix that maximizes the statistical independence between rows of the output signal matrix (i.e., each independent component) when multiplied by the original signal matrix. After separation, the unit extracts features from each separated independent component (IC), calculating its time-series kurtosis (for identifying spike-shaped EMG artifacts), power spectral density (for identifying power interference artifacts with specific frequency peaks), and its spatial projection on the scalp (for identifying blink artifacts with typical forehead distribution). Finally, by comparing these extracted features with a pre-defined artifact feature template library, or using a pre-trained classifier, each independent component is labeled with a brain signal or a specific type of artifact, ultimately outputting a list of components identified as artifacts.

[0085] The artifact removal unit 1022 is used to filter out artifact components according to a preset artifact removal threshold.

[0086] Specifically, this unit executes immediately after the artifact recognition unit 1021 completes its labeling work. The execution process is as follows: First, the unit receives the unmixing matrix, all separated independent components, and a list labeling which components are artifacts. Next, for each independent component labeled as an artifact in the list, a filtering operation is performed according to a preset artifact removal threshold. After processing all specified artifact components, this unit multiplies the original mixing matrix (the inverse of the unmixing matrix) by this modified (some components have been set to zero or attenuated) independent component matrix, thereby reconstructing a multichannel EEG data free from artifact influence.

[0087] The data quality assessment unit 1023 is used to score the quality of preprocessed EEG data based on signal-to-noise ratio, spectral characteristics, and signal integrity.

[0088] Specifically, this unit executes after the artifact removal unit 1022 completes data cleansing. Its workflow is as follows: First, the unit segments the cleaned data. For each segment, it calculates multiple quality metrics. For example, it estimates the signal-to-noise ratio by calculating the ratio of the average power in a specific frequency band (e.g., 1-40Hz) to the average power in a higher frequency band (e.g., 70-100Hz, typically considered noise). Simultaneously, it calculates the power spectrum of each channel, detects the presence of expected physiological peaks (e.g., alpha waves), and fits the spectral slope. Furthermore, it scans the time series of each channel, statistically analyzing the percentage of data points with zero or constant signal values ​​to assess signal integrity. Finally, the calculation results for all channels and all metrics are merged into a final quality score based on preset weights (e.g., the weight for the alpha wave in the pillow region can be set higher).

[0089] The preprocessing parameter optimization unit 1024 is used to adjust the preset artifact removal threshold based on the data quality score results.

[0090] Specifically, this unit is triggered after the data quality assessment unit 1023 provides a score. Its workflow is as follows: Upon receiving a quality score lower than a preset quality score threshold, if the report indicates that the physiological characteristics of the signal (such as alpha waves) have become blurred, it means the artifact removal process is too aggressive. The optimization unit then decides to increase the preset artifact removal threshold, making it more conservative. After determining the new preset artifact removal threshold, it returns to the starting point of the preprocessing process and uses this new preset threshold to perform a complete round of preprocessing on the original data. This processing-evaluation-tuning-reprocessing cycle continues until the quality score exceeds the preset quality score threshold or the preset maximum number of iterations is reached.

[0091] Please see Figure 3This is a schematic diagram of the functional network analysis module of the high-precision transcranial direct current stimulation device in this application embodiment.

[0092] Functional network analysis module 103 includes: The EEG signal extraction unit 1031 is used to extract EEG signals of a preset frequency band from EEG data. The preset frequency band includes at least delta waves, theta waves, alpha waves, beta waves and gamma waves.

[0093] Specifically, this unit executes after the data preprocessing module completes data purification. Its workflow is as follows: this unit applies a set of digital bandpass filters in parallel to the preprocessed EEG signals of each channel. Each filter is precisely designed to allow only a preset frequency band of signal to pass through. For example, an alpha wave filter will filter out all signal components with frequencies below 8 Hz and above 13 Hz, retaining only signals within this range. To prevent the filtering process from introducing time delays that could affect subsequent phase analysis, zero-phase filters are typically used (e.g., through forward and backward filtering). After this series of parallel filtering operations, each input single-channel broadband signal is decomposed into multiple parallel single-channel narrowband signal streams corresponding to different frequency bands, such as δ, θ, α, β, and γ, for subsequent units to analyze separately.

[0094] The spectrum analysis unit 1032 is used to perform frequency domain conversion on EEG data using fast Fourier transform, calculate the power spectral density of each electrode position in each frequency band, and normalize to obtain the power ratio of each EEG wave at each electrode position.

[0095] Specifically, this unit executes after the EEG signal is extracted into various frequency bands (or directly analyzes the broadband signal). The process is as follows: First, the EEG signal of each channel is divided into multiple overlapping data segments. Second, a window function (such as the Hanning window) is applied to each data segment to reduce spectral leakage. Then, an FFT is performed on each windowed data segment to calculate its power spectrum. Finally, the power spectra of all data segments are averaged to obtain the average power spectral density (PSD) of that channel. Based on this PSD, the absolute power of these EEG waves can be calculated by integrating over the frequency ranges of each preset frequency band (δ, θ, α, β, γ). Dividing the absolute power of each frequency band by the sum of the absolute powers of all frequency bands yields the power percentage of that frequency band. This process is repeated for all electrode locations.

[0096] The functional connectivity calculation unit 1033 is used to calculate the functional connectivity strength between brain regions in each frequency band using coherence analysis methods, and to construct a multi-band functional connectivity network.

[0097] Coherence analysis is a frequency domain analysis technique used to quantify the linear correlation between two time series at a specific frequency, with values ​​ranging from 0 (completely uncorrelated) to 1 (completely linearly correlated). Functional connectivity strength, calculated using methods such as coherence analysis, represents the synchronicity of neural activity between two brain regions. A multi-band functional connectivity network is a collection of multiple networks, each corresponding to a frequency band. Nodes in the network represent brain regions (electrodes), and the weights of the connections (edges) between nodes are defined by the functional connectivity strength of that frequency band.

[0098] Specifically, this unit executes after the EEG signal is decomposed into various frequency bands. Its workflow is as follows: Taking the alpha band as an example, the unit traverses all possible electrode pairs in the brain (e.g., for 16 electrodes, there are 120 unique electrode pairs). For each electrode pair (e.g., F3-P3), the unit extracts their respective alpha band signals. Then, it calculates the cross-power spectral density and the individual auto-power spectral density of these two signals. Finally, according to the formula: coherence = |cross-power spectral density|^2 / (auto-power spectral density of signal 1 × auto-power spectral density of signal 2), the coherence value of F3 and P3 in the alpha band is calculated. This value serves as the functional connectivity strength between F3 and P3 in the alpha band. The calculation results for all electrode pairs are filled into a 16×16 symmetric matrix, thus constructing the functional connectivity network of the alpha band. This process is repeated for all other frequency bands such as δ, θ, β, and γ, ultimately generating a multi-band functional connectivity network.

[0099] The deviation calculation unit 1034 is used to statistically compare the power ratio and multi-band functional connectivity network of the subjects with the preset standard EEG atlas and preset functional network patterns of the same age group, and calculate the deviation.

[0100] Specifically, this unit executes after spectral analysis and functional connectivity calculations are completed. Its core mechanism is as follows: This unit receives complete power percentage and functional connectivity strength data from the subject. For each indicator (e.g., alpha power percentage at the O1 electrode, or functional connectivity strength of F3-P4 in the beta band), it queries a large, age-stratified standardized database. The database returns the mean and standard deviation of that indicator in the healthy population of the same age group as the subject. Then, the unit uses a Z-score calculation method to calculate the subject's indicator value against the population mean and standard deviation, thus converting each raw indicator into a standardized deviation score. This process systematically covers all power and connectivity indicators, ultimately generating a comprehensive deviation score matrix that reflects the deviation of the subject's various brain function indicators from the normal range.

[0101] The abnormal functional subnetwork identification unit 1035 is used to identify combinations of brain regions with abnormal power distribution and abnormal connection patterns based on deviation, and to determine the abnormal functional subnetwork.

[0102] Specifically, this unit executes after the deviation calculation is complete. Its workflow is as follows: First, the unit receives the comprehensive deviation score matrix. Then, a significance level threshold is set, which typically corresponds to a statistical p-value (e.g., setting the absolute value of the deviation score greater than 1.96 corresponds to a two-sided p < 0.05). Next, all deviation scores are traversed: all electrode locations with power percentage deviation scores exceeding the threshold are marked as anomalous nodes, and all connections with functional connectivity strength deviation scores exceeding the threshold are marked as anomalous edges. Finally, the graph structure formed by all these marked anomalous nodes and edges is defined as the subject's anomalous functional subnetwork. This subnetwork visually demonstrates the core of the subject's brain functional abnormalities.

[0103] Please see Figure 4 This is a schematic diagram of the deviation calculation unit of the high-precision transcranial direct current stimulation device in the embodiments of this application.

[0104] Deviation calculation unit 1034 includes: The power percentage standard score calculation subunit 10341 is used to calculate the power percentage of each frequency band at each electrode position and the power percentage standard score of the preset standard EEG spectrum for the same age group.

[0105] Specifically, this subunit executes after the spectral analysis is completed and the subject's power percentage data is obtained. Its execution process is as follows: First, the subunit receives a matrix representing the subject's power percentage (e.g., 16 electrodes × 6 frequency bands). Then, for each element in the matrix (e.g., the alpha wave power percentage at the O1 electrode location), it queries a standard EEG database and, based on the subject's age, extracts the mean (μ) and standard deviation (σ) of the power percentage in the normal population corresponding to that specific location and frequency band. Next, it applies the Z-score calculation formula: Z = (subject observation value - population mean) / population standard deviation. This calculation generates a corresponding Z-score for each element in the matrix. Finally, the subunit outputs a matrix with the same dimensions as the input matrix, but with all numerical content converted to power percentage standard scores.

[0106] The functional connectivity standard score calculation subunit 10342 is used to calculate the functional connectivity standard score of each frequency band in the brain region and the corresponding connection of the preset standard functional network pattern of the same age group.

[0107] Specifically, this subunit is executed after functional connectivity analysis is completed and the functional connectivity strength data of the subjects is obtained. Its execution process is as follows: First, the subunit receives a set of matrices representing the functional connectivity strength of the subjects (e.g., a 16×16 connectivity matrix for each frequency band). Then, it iterates through all unique brain region connections in each matrix (e.g., connections between Fp1 and Fp2, connections between Fp1 and F7, etc.). For each specific connection (e.g., the connection strength of P3-P4 in the β band), it queries the standard functional network pattern database to extract the mean (μ) and standard deviation (σ) of the normal population corresponding to that specific connection in that frequency band, based on the subject's age. Next, it calculates the Z-score for that connection using the standard score formula. This process calculates a corresponding Z-score for all brain region connections and all frequency bands, ultimately outputting a matrix with the same dimensions as the input matrix, but with all numerical content converted to functional connectivity standard scores.

[0108] The comprehensive deviation matrix subunit 10343 is used to combine the power ratio standard score and the functional connection standard score to form a comprehensive deviation matrix, and the comprehensive deviation matrix is ​​used as the deviation.

[0109] Specifically, this sub-unit is the final step in the deviation calculation process, executed after the standard scores for power proportion and functional connectivity have been calculated. Its execution process is as follows: First, the sub-unit receives the standard score matrix for power proportion (e.g., 16×6 dimensional) and the standard score matrix for functional connectivity (e.g., six 16×16 dimensional matrices). Next, it performs a predefined "flattening and splicing" operation. For example, it first flattens the 16×6 standard score matrix for power proportion by column or row, forming a vector of length 96. Then, for each standard score matrix for functional connectivity, it extracts the non-repeating connectivity scores from the upper or lower triangular portion (120 non-repeating connections for 16 electrodes), flattens and splices the connectivity scores for all six frequency bands, forming a vector of length 720 (120×6). Finally, it concatenates these two vectors (the power deviation vector of length 96 and the connectivity deviation vector of length 720) end-to-end, forming a longer single vector of length 816. This long vector, which contains all the deviation information, is defined as the comprehensive deviation matrix (or vector).

[0110] Please see Figure 5 This is a schematic diagram of the individual difference adaptation module of the high-precision transcranial direct current stimulation device in the embodiments of this application.

[0111] Individual Difference Adaptation Module 106 includes: The head geometric feature extraction unit 1061 is used to measure geometric parameters from the subject's brain structure imaging data. The geometric parameters include at least head circumference, skull thickness, cerebrospinal fluid thickness, and gray matter volume.

[0112] Specifically, this unit executes after the navigation and positioning module has constructed a preliminary 3D model and before the electric field simulation is performed. Its workflow is as follows: First, the unit receives the individual's MRI data and invokes automated segmentation algorithms from mature neuroimaging processing software (such as FreeSurfer, SPM, etc.) to classify each voxel in the 3D image into different tissues, precisely delineating the boundaries of different tissues such as the scalp, skull, cerebrospinal fluid, gray matter, and white matter. Based on these segmentation results, the unit performs a series of measurements: calculating the head circumference by extracting the contours of specific planes on the scalp surface model; obtaining the skull thickness by calculating the average distance between the inner and outer surfaces of the skull; similarly calculating the cerebrospinal fluid thickness; and calculating the total gray matter volume by counting all voxels labeled as gray matter and multiplying by the volume per voxel. Finally, the unit outputs a set of quantitative geometric parameters that accurately describe the unique anatomical morphology of the subject's head.

[0113] The individualized conductivity calculation unit 1062 is used to calculate the conductivity of each layer of the subject's brain tissue based on the statistical relationship between age and brain tissue conductivity.

[0114] Specifically, this unit executes after extracting the head's geometric features and before performing the electric field simulation. The execution process is as follows: First, the unit receives the subject's exact age. Then, it accesses a built-in parameter database that stores the functional relationship between the electrical conductivity of different tissues and age. For skull tissue, where conductivity changes most significantly with age, the unit substitutes the subject's age into a pre-defined regression equation (e.g., σ_skull = -0.0001*age + 0.018) to calculate an age-appropriate skull conductivity value. For other tissues, such as the scalp, cerebrospinal fluid, gray matter, and white matter, whose conductivity changes less with age, this unit directly uses widely accepted standard values ​​from the literature, or similarly uses an age-based fine-tuning function. Finally, the unit outputs a complete set of tissue conductivity parameters tailored to the subject.

[0115] The electric field conduction efficiency evaluation unit 1063 is used to calculate the electric field conduction efficiency coefficient from the electrodes to each target brain region based on individualized conductivity and head geometry using the finite element method.

[0116] The Finite Element Method (FEM) is a powerful numerical computation technique that discretizes complex continuous geometries (such as the head) into millions of tiny, regularly shaped units (such as tetrahedrons). The physical field distribution of the entire object is then approximated by solving the physical equations (in this case, electrical equations) for each unit. The target brain region refers to the specific cortical area where the abnormal functional subnetwork, identified by the functional network analysis module, is located, serving as a neuromodulation target. The electric field conduction efficiency coefficient is a key quantitative indicator used to represent the intensity of the electric field (in volts per meter) generated per unit input current (e.g., per milliampere) in the target brain region, directly reflecting the conductivity of the individual head structure to the stimulating current.

[0117] Specifically, this unit executes after the initial stimulation protocol is generated and before the final parameter adjustments. Its detailed process is as follows: First, this unit integrates the outputs of the previous two units, constructing a high-precision finite element mesh model of the head based on the individual's geometric characteristics, and assigning corresponding individualized conductivity values ​​to units belonging to different tissues within the model. Then, the electrode positions and current intensities defined in the initial stimulation protocol are used as boundary conditions for the model. Next, the finite element solver is activated to calculate the potential distribution throughout the head model under these conditions. By calculating the negative gradient of the potential distribution, a three-dimensional electric field vector field is obtained. Finally, the unit uses the target brain region as the region of interest (ROI), calculates the average electric field intensity within this region, and divides this average value by the total applied current to obtain the final electric field conduction efficiency coefficient.

[0118] The stimulation parameter individualization adjustment unit 1064 is used to individualize the multi-target synergistic stimulation scheme output by the preset machine learning model according to the electric field conduction efficiency coefficient, and to increase the current intensity of individuals with conduction efficiency less than the preset efficiency threshold by a preset adjustment ratio.

[0119] Specifically, this unit executes immediately after the electric field conduction efficiency is evaluated. Its workflow is as follows: First, the unit receives the calculated individual electric field conduction efficiency coefficient and compares it with a preset efficiency threshold. If the individual's efficiency coefficient is greater than or equal to the threshold, it indicates that the preliminary scheme is expected to generate a sufficiently strong electric field on that individual, therefore the scheme is approved without modification. Conversely, if the individual's efficiency coefficient is less than the threshold, the unit initiates an adjustment procedure. A current gain factor is calculated based on a preset adjustment ratio (e.g., a simple inverse relationship: adjusted current = original current × (efficiency threshold / individual efficiency coefficient)). Then, the current intensity of all electrodes in the preliminary scheme is multiplied by this gain factor.

[0120] Please see Figure 6This is a schematic diagram of the electric field conduction efficiency evaluation unit of the high-precision transcranial direct current stimulation device in this application embodiment.

[0121] The electric field conduction efficiency evaluation unit 1063 includes: Finite element mesh generation sub-element 10631 is used to divide the three-dimensional navigation model into tetrahedral mesh elements with a side length not greater than the preset mesh size.

[0122] Specifically, this sub-unit executes immediately after obtaining a 3D navigation model composed of multiple closed surfaces representing the individual's anatomical structure. Its workflow is as follows: First, the sub-unit receives surface mesh data describing the boundaries of tissues such as the scalp, skull, and brain. Then, it invokes a mesh generation algorithm (such as the constrained Delaunay triangulation algorithm) to perform a filling-based meshing of the volume space enclosed by these surfaces. During the meshing process, the algorithm adheres to preset mesh size constraints, ensuring that the longest side of any generated tetrahedral element does not exceed this size. Furthermore, to ensure the stability of numerical computation, the algorithm optimizes the mesh quality, for example, by avoiding the generation of overly flat or elongated distorted elements. Finally, the sub-unit outputs a volumetric mesh model composed of millions of tetrahedral elements that fully describes the multi-layered structure of the head, with each element labeled with its corresponding tissue type.

[0123] The boundary condition setting subunit 10632 is used to map the electrode positions generated by the transcranial direct current stimulation scheme module onto the tetrahedral mesh units constituting the scalp surface as voltage boundary conditions, and to set other areas of the scalp as zero current density boundary conditions. Other areas are areas other than the electrode positions.

[0124] Specifically, this sub-element executes after the finite element mesh is generated but before the solver is started. Its execution process is as follows: First, based on the electrode center and dimensions determined by the navigation and positioning module, the sub-element identifies all tetrahedral mesh elements on the scalp surface mesh that are covered by the anode and cathode electrodes. Then, it assigns a preset potential value (e.g., +V) to all tetrahedral mesh elements belonging to the anode and another potential value (e.g., -V or 0) to all tetrahedral mesh elements belonging to the cathode. For all other tetrahedral mesh elements on the scalp surface that do not belong to any electrode, the sub-element applies a constraint of zero normal current density. These predefined boundary conditions, together with the Laplace equation describing the internal physical laws of the model, constitute a complete mathematical problem with a unique solution.

[0125] The potential distribution solution sub-element 10633 is used to solve the Laplace equation using the conjugate gradient method. It calculates the potential value of each grid node by means of the constraint that the divergence between conductivity and potential gradient is zero.

[0126] Specifically, this sub-element is executed after mesh generation and boundary condition setting. Its workflow is as follows: The finite element method first transforms the Laplace equation acting on each tiny tetrahedral element into a linear algebraic equation. Then, through an assembly process, the equations of millions of elements are combined into a huge linear system of equations of the form [K]{V}={F}. Here, [K] is the global stiffness matrix, determined by the mesh geometry and microstructure conductivity. {V} is the unknown vector containing the potential values ​​of all nodes to be solved. {F} is the load vector determined by the boundary conditions. Since the [K] matrix is ​​enormous, direct inversion is not feasible. Therefore, the conjugate gradient method is used, starting from an initial guessed solution and iteratively searching along a series of mutually conjugate directions. Each step makes the solution closer to the true value until the difference between two iterations is less than a very small convergence tolerance. Finally, this sub-element outputs the converged vector {V}, which represents the precise potential values ​​of all nodes.

[0127] The conduction efficiency calculation subunit 10634 is used to calculate the conduction efficiency coefficient by the ratio of the electric field gradient magnitude of the abnormal function subnetwork to the electrode current.

[0128] Specifically, this sub-unit is executed after the potential distribution solution is completed. Its workflow is as follows: First, the sub-unit receives a vector {V} containing the potential values ​​of all nodes. Then, it identifies all mesh nodes located within the predefined anomalous functional sub-network. For each tetrahedral element within the anomalous functional sub-network, the approximate uniform electric field vector E within that element is obtained by calculating the spatial gradient of the potential values ​​at its four vertices. Next, the amplitude (i.e., magnitude |E|) of the electric field vector E for each element is calculated. Subsequently, the electric field amplitudes of all elements within the anomalous functional sub-network are weighted and averaged (usually weighted by element volume) to obtain a value representing the average electric field strength of the anomalous functional sub-network. Finally, this average electric field strength value is divided by the total stimulation current set in the boundary conditions to obtain the final electric field conduction efficiency coefficient.

[0129] Please see Figure 7 This is a schematic diagram of the stimulation safety protection module of the high-precision transcranial direct current stimulation device in this application embodiment.

[0130] Stimulation safety protection module 108 includes: The current density monitoring unit 1081 is used to monitor the current density of the skin beneath each electrode.

[0131] The impedance change detection unit 1082 is used to measure the impedance of the electrode in contact with the skin at a preset detection frequency. When the impedance change value is greater than the preset change threshold, it is determined that the electrode has fallen off or has poor contact.

[0132] The temperature monitoring unit 1083 is used to monitor changes in skin temperature at the electrode site. When the temperature exceeds a first preset temperature threshold, the current intensity is reduced according to a preset adjustment ratio.

[0133] The emergency stop execution unit 1084 is used to cut off the power supply to all electrodes within a preset response time when the detected current density is greater than a preset safe current density threshold, the impedance change value is greater than a preset change threshold, or the temperature is greater than a second preset temperature threshold.

[0134] The security incident recording unit 1085 is used to record information about each security incident and establish a security file.

[0135] Please see Figure 8 This is a schematic diagram of the change calculation module of the high-precision transcranial direct current stimulation device in this application embodiment.

[0136] The change calculation module 109 includes: The post-stimulation data acquisition unit 1091 is used to acquire the subject's electroencephalogram (EEG) data at a preset time point after the stimulation ends.

[0137] The corresponding analysis and processing unit 1092 processes the post-stimulation EEG data through the same digital filtering, spectral analysis, and functional connectivity analysis process as before stimulation.

[0138] Specifically, this unit executes after post-stimulation data acquisition is complete. Its workflow is as follows: First, the unit retrieves all parameter configuration files used in processing the subject's pre-stimulation data from the system. This file details the specific settings for each step in the data preprocessing module 102 and the functional network analysis module 103. Then, the unit applies this parameter configuration file to the newly acquired post-stimulation EEG data. Its output is analysis results that are completely consistent with the pre-stimulation data in terms of format, dimensions, and indicator types; for example, a post-stimulation power proportion matrix and a set of post-stimulation multi-band functional connectivity network matrices. These two results can now be directly compared point-by-point and connection-by-connection with the pre-stimulation results.

[0139] The band power change calculation unit 1093 is used to calculate the difference in power ratio before and after stimulation on an electrode-by-electrode and frequency-by-frequency-band basis, and to quantify the degree of power change in each brain region and frequency band.

[0140] Specifically, this unit is executed after the corresponding analysis and processing unit 1092 generates two sets of power spectrum data before and after stimulation. Its workflow is as follows: This unit receives two power proportion matrices of the same dimension: P_pre (number of electrodes × number of frequency bands) and P_post (number of electrodes × number of frequency bands). Then, it performs a simple matrix subtraction operation: ΔP = P_post - P_pre. Each element in the resulting difference matrix ΔP represents the change in the power proportion of a specific frequency band on a specific electrode. For example, a positive ΔP(i,j) indicates that the EEG activity of the j-th frequency band on the i-th electrode was relatively enhanced after stimulation. A negative value indicates a decrease. This unit ultimately outputs this difference matrix, providing direct evidence of changes in localized brain activity for subsequent comprehensive assessment.

[0141] The network connectivity change calculation unit 1094 is used to calculate the strength changes of corresponding connections in the functional connectivity network before and after stimulation, and to identify connection patterns where the connection strength is enhanced, weakened, or restored to normal.

[0142] Specifically, this unit executes after the corresponding analysis and processing unit 1092 generates two sets of functional connectivity network matrices before and after stimulation. Its workflow is as follows: For each frequency band, this unit receives two functional connectivity matrices, C_pre and C_post. First, it calculates the intensity change matrix using matrix subtraction ΔC = C_post - C_pre. Then, it classifies each element in ΔC: if the value is greater than a small positive threshold, it is marked as enhanced; if it is less than a negative threshold, it is marked as weakened. For identification of recovery to normal, this unit also retrieves the subject's pre-stimulation deviation matrix. If a connection has a high deviation score before stimulation (e.g., |Z|>2, indicating a significant abnormality), but the post-stimulation connection strength value falls near the mean of the normal population distribution (e.g., post-stimulation deviation score |Z|<1), then that connection is specifically marked as recovered to normal.

[0143] The integrated change unit 1095 is used to integrate band power changes and network connectivity changes into a comprehensive evaluation index of neuromodulation effects.

[0144] Specifically, this unit executes after all specific changes have been calculated. Its workflow is as follows: the unit receives the power change matrix ΔP and the connectivity change matrix ΔC. One integration approach is as follows: first, nodes and connections identified as anomalous functional subnetworks before stimulation are identified. Then, a targeting effect score is calculated, which is a weighted sum of the degree to which power and connectivity in these target regions return to normal. For example, a connection that was anomalously enhanced but weakened after stimulation would receive a positive score. Simultaneously, a non-target area impact score is calculated to assess potential unintended side effects. Finally, these scores can be combined into a single comprehensive evaluation index or presented to the operator in dashboard form.

[0145] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, 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. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0146] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0147] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0148] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A high-precision transcranial direct current stimulation device based on a navigation system, characterized in that, include: The EEG data acquisition module is used to collect the EEG data of the subject. The functional network analysis module performs spectral analysis and functional connectivity analysis on the EEG data. The spectral analysis obtains the power percentage of each EEG wave at each electrode position. The functional connectivity analysis constructs a functional connectivity network by calculating the functional connectivity strength between brain regions. The power percentage and the functional connectivity network are compared with preset standard EEG atlases and functional network patterns for the same age group to identify abnormal functional subnetworks and calculate the deviation of the power percentage and the functional connectivity network. The navigation and positioning module is used to determine the location of multi-target transcranial direct current stimulation based on the three-dimensional navigation model constructed from the individual brain structure imaging data of the subject and the abnormal functional subnetwork. The transcranial direct current stimulation scheme generation module includes a preset machine learning model, which is used to generate a multi-target synergistic stimulation scheme by taking the subject's age, the deviation degree and the multi-target transcranial direct current stimulation location as input features. A transcranial direct current stimulation execution module is used to implement the multi-target synergistic stimulation protocol on the subject.

2. The high-precision transcranial direct current stimulation device according to claim 1, characterized in that, The transcranial direct current stimulation protocol generation module specifically includes: The decision tree integration construction unit is used to construct a gradient boosting decision tree model based on the subject's age, deviation, multi-target transcranial direct current stimulation location, and effective historical multi-target synergistic stimulation schemes. The gradient boosting training unit is used to perform iterative training by using the negative gradient as an approximation of the residual. In each iteration, a new decision tree is trained to fit the residual generated by the previous prediction. The feature importance assessment unit is used to determine the subject's age, the degree of deviation, and the weight contribution of the multi-target transcranial direct current stimulation location in the prediction by calculating the number of splits and information gain of each input feature in all decision trees. The stimulation parameter output unit is used to sum the prediction results of all decision trees according to a preset learning rate and output a multi-target synergistic stimulation scheme consisting of the current intensity, polarity, and stimulation duration parameters of each electrode.

3. The high-precision transcranial direct current stimulation device according to claim 1, characterized in that, It also includes an individual difference adaptation module, including: A head geometric feature extraction unit is used to measure geometric parameters from the brain structure imaging data of the subject, the geometric parameters including at least head circumference, skull thickness, cerebrospinal fluid thickness and gray matter volume; A personalized conductivity calculation unit is used to calculate the conductivity of each layer of the subject's brain tissue based on the statistical relationship between age and brain tissue conductivity. The electric field conduction efficiency evaluation unit is used to calculate the electric field conduction efficiency coefficient from the electrodes to each target brain region based on the conductivity of each layer of the subject's brain tissue and the geometric features of the head using the finite element method. The stimulation parameter individualization adjustment unit is used to individualize the multi-target synergistic stimulation scheme output by the preset machine learning model according to the electric field conduction efficiency coefficient, and to increase the current intensity of individuals with conduction efficiency less than the preset efficiency threshold by a preset adjustment ratio.

4. The high-precision transcranial direct current stimulation device according to claim 3, characterized in that, The electric field conduction efficiency evaluation unit includes: Finite element mesh generation sub-units are used to divide the three-dimensional navigation model into tetrahedral mesh units with side lengths no greater than a preset mesh size; The boundary condition setting subunit is used to map the electrode position generated by the transcranial direct current stimulation scheme module onto the tetrahedral mesh unit constituting the scalp surface as a voltage boundary condition, and set other areas of the scalp as zero current density boundary conditions, wherein the other areas are areas other than the electrode positions. The potential distribution solution sub-element is used to solve the Laplace equation using the conjugate gradient method. It calculates the potential value of each grid node by means of the constraint that the divergence between conductivity and potential gradient is zero. The conduction efficiency calculation subunit is used to calculate the conduction efficiency coefficient by the ratio of the electric field gradient magnitude of the abnormal function subnetwork to the electrode current.

5. The high-precision transcranial direct current stimulation device according to claim 1, characterized in that, It also includes a data preprocessing module, including: The artifact detection unit is used to identify artifact components of the EEG data using an independent component analysis method. The artifact components include at least artifact components caused by blinking, electromyography, and power supply interference. The artifact removal unit is used to filter out the artifact components according to a preset artifact removal threshold. The data quality assessment unit is used to score the quality of preprocessed EEG data based on signal-to-noise ratio, spectral characteristics, and signal integrity. The preprocessing parameter optimization unit is used to adjust the preset artifact removal threshold based on the data quality score.

6. The high-precision transcranial direct current stimulation device according to claim 1, characterized in that, The functional network analysis module includes: The EEG signal extraction unit is used to extract EEG signals of a preset frequency band from the EEG data, wherein the preset frequency band includes at least delta waves, theta waves, alpha waves, beta waves and gamma waves; The spectrum analysis unit is used to perform frequency domain conversion on the EEG data using fast Fourier transform, calculate the power spectral density of each electrode position in each frequency band, and normalize it to obtain the power ratio of each EEG wave at each electrode position. The functional connectivity computing unit is used to calculate the functional connectivity strength between brain regions in each frequency band using coherence analysis methods, and to construct a multi-band functional connectivity network. The deviation calculation unit is used to statistically compare the power ratio and multi-band functional connectivity network of the subject with the preset standard EEG atlas and preset functional network pattern of the same age group, and calculate the deviation. An abnormal functional subnetwork identification unit is used to identify combinations of brain regions with abnormal power distribution and abnormal connection patterns based on the deviation, and to determine the abnormal functional subnetwork.

7. The high-precision transcranial direct current stimulation device according to claim 6, characterized in that, The deviation calculation unit includes: The power percentage standard score calculation subunit is used to calculate the power percentage of each frequency band at each electrode position and the power percentage standard score of the preset standard EEG spectrum for the same age group. The functional connectivity standard score calculation subunit is used to calculate the functional connectivity standard score of each frequency band in the brain region and the connection corresponding to the preset standard functional network pattern of the same age group. The comprehensive deviation matrix sub-unit is used to combine the power ratio standard score and the functional connection standard score to form a comprehensive deviation matrix, and to use the comprehensive deviation matrix as the deviation.

8. The high-precision transcranial direct current stimulation device according to claim 1, characterized in that, It also includes a change calculation module, which includes: The post-stimulation data acquisition unit is used to acquire the subject's electroencephalogram (EEG) data at a preset time point after the stimulation ends. The corresponding analysis and processing unit processes the post-stimulation EEG data through the same digital filtering, spectral analysis, and functional connectivity analysis process as before stimulation. The band power change calculation unit is used to calculate the difference in power ratio before and after stimulation on an electrode-by-electrode and frequency-by-frequency band basis, and to quantify the degree of power change in each brain region and frequency band. The network connectivity change calculation unit is used to calculate the strength changes of corresponding connections in the functional connectivity network before and after stimulation, and to identify connection patterns that are strengthened, weakened or restored to normal. The integrated change unit is used to integrate band power changes and network connectivity changes into a comprehensive evaluation index of neuromodulation effects.

9. The high-precision transcranial direct current stimulation device according to claim 8, characterized in that, It also includes a feedback optimization module, which is used to update the machine learning model using the neural modulation effect evaluation index as feedback data.

10. The high-precision transcranial direct current stimulation device according to claim 1, characterized in that, It also includes a safety protection module, including: A current density monitoring unit is used to monitor the current density of the skin beneath each electrode; The impedance change detection unit is used to measure the impedance of the electrode in contact with the skin at a preset detection frequency. When the impedance change value is greater than the preset change threshold, it is determined that the electrode has fallen off or has poor contact. The temperature monitoring unit is used to monitor changes in skin temperature at the electrode site. When the temperature exceeds a first preset temperature threshold, the current intensity is reduced according to a preset adjustment ratio. An emergency stop execution unit is used to cut off the power supply to all electrodes within a preset response time when the current density is detected to be greater than the preset safe current density threshold, the impedance change value is greater than the preset change threshold, or the temperature is greater than the second preset temperature threshold. The security incident recording unit is used to record information about each security incident and establish a security file.

Images