An intelligent transcranial micro-current stimulation method and system for neuromodulation

Abstract

The application discloses an intelligent transcranial micro-current stimulation method and system for neuromodulation, and relates to the field of information technology, wherein electroencephalogram signal data is acquired through a dry electrode array, and biomarker features are extracted based on Fourier transform; a personalized micro-current stimulation waveform is generated according to the biomarker features, the waveform is dynamically varied to simulate brain wave chaos and avoid neural adaptive tolerance; a craniocerebral electric field model is generated through multi-physical field coupling modeling, skin impedance data and electroencephalogram feedback signals are integrated, a controller is used to dynamically adjust electrode output, and current is delivered to a target brain area; electroencephalogram data is processed based on a time series analysis model, precursors of emotional troughs are predicted, an intervention program is triggered, and an emotional map is generated to optimize a conditioning scheme. The application adopts the above-mentioned intelligent transcranial micro-current stimulation method and system for neuromodulation, realizes personalized and accurate emotional regulation and neural stimulation, and significantly improves long-term effectiveness and user experience.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an intelligent transcranial microcurrent stimulation method and system for neural modulation. Background Technology

[0002] Neuroscience and emotion management play a crucial role in improving mental health by modulating brain activity. Electroencephalography (EEG), as a key indicator of brain activity, can reveal the underlying mechanisms of emotional fluctuations and is therefore widely used to assist in emotion regulation and psychological intervention. However, current technologies face numerous challenges in practical applications, struggling to meet the needs of personalized and precise interventions, particularly in achieving continuous and effective neuromodulation in dynamic environments. This not only affects intervention effectiveness but also limits the improvement of user experience. Current emotion management devices typically rely on fixed-pattern electrical stimulation, making it difficult to adapt to individual differences. Many devices intervene in the brain through preset stimulation parameters, but these parameters are often based on group averages, ignoring the unique EEG activity patterns of individual users. For example, a user may experience abnormal activity in specific brain regions due to stress, and a fixed stimulation pattern cannot specifically modulate that area, resulting in poor efficacy. Furthermore, individual differences in skin impedance and brain tissue structure affect the efficiency of current transmission; existing methods lack dynamic adjustment mechanisms, easily leading to insufficient or excessive stimulation intensity. A deeper challenge lies in how to maintain synchronization with the dynamic activity of the user's brain. The neural oscillations of the brain are highly dynamic, and emotional fluctuations can change rapidly within a short period. For example, users may experience mood swings throughout the day due to work stress or unexpected events, and existing devices struggle to capture these changes in real time and adjust stimulation protocols accordingly. Traditional signal processing methods often fail to accurately separate key biomarkers when extracting EEG signal features, leading to a disconnect between stimulation waveforms and brain activity, thus reducing the effectiveness of interventions. Therefore, how to generate dynamic stimulation waveforms that are phase-synchronized with the user's brain activity in real time and ensure precise delivery of current to the target brain region in complex physiological environments has become a key issue in the field of emotion management technology. Furthermore, the dynamic nature of emotion management is not only reflected in the rapid changes in EEG activity but also in how to accurately predict the trend of emotional fluctuations before intervention. Mood swings often have specific precursor signals, such as abnormal enhancements in certain frequency bands of EEG, but existing technologies struggle to accurately identify and predict these precursors when processing these complex time-series signals. For example, a user may experience mild EEG signal abnormalities due to nervousness before a meeting, but existing devices typically only intervene after the emotional problem has manifested, missing the optimal window for prevention. This lag significantly reduces the effectiveness of interventions, especially in scenarios requiring sustained emotional stability, such as long-term psychological stress management.

[0003] Another key challenge lies in the complexity of human physiology, which limits the precise delivery of electrical stimulation. Skin impedance varies constantly due to factors such as sweat and temperature, while the non-uniformity of brain tissue makes the current path unpredictable. For example, the same stimulation parameters may result in inconsistent current intensities reaching the target brain region among different users due to differences in scalp thickness and brain tissue density. Existing devices typically lack the ability to monitor and dynamically compensate for these physiological variables in real time, making it difficult to ensure that each stimulus is precisely applied to the target area. This instability further complicates personalized adjustments, making it difficult to sustain the effects of long-term interventions.

[0004] Therefore, how to generate stimulation waveforms that are synchronized with the user's brain activity in a dynamically changing physiological environment, combined with real-time EEG signal analysis, and how to ensure that the current is accurately applied to the target brain region through dynamic adjustment, has become a key issue in the field of emotion management technology. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent transcranial microcurrent stimulation method and system for neural modulation, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent transcranial microcurrent stimulation method for neural modulation, comprising: S1. Obtain EEG signal data through a dry electrode array and extract biomarker features based on Fourier transform; S2. Generate personalized microcurrent stimulation waveforms based on biomarker characteristics. These waveforms dynamically vary to simulate brainwave chaos and avoid neural adaptive tolerance. The waveforms are phase-synchronized with brain electrical oscillations. S3. A brain electric field model is generated through multi-physics coupling modeling, skin impedance data and EEG feedback signals are integrated, and the electrode output is dynamically adjusted by a controller to deliver current to the target brain region. S4. Based on time series analysis models, EEG data is processed to predict early signs of emotional lows and trigger intervention procedures. At the same time, an emotion map is generated to optimize subsequent treatment plans.

[0007] Preferably, step S1 specifically includes: A dry electrode array was used to acquire multi-channel EEG signal data, and signal preprocessing and filtering were applied during the acquisition process to remove noise interference and obtain a clean EEG signal sequence. The system inputs the clean EEG signal sequence into the Fourier transform module, calculates the power spectral density, and extracts the relative ratio of alpha and theta waves as a biomarker feature. If the relative ratio exceeds a preset threshold, it is determined to be an emotional fluctuation state and the corresponding timestamp is marked. Feature vectors are constructed based on biomarker features, and time-domain and frequency-domain indicators are fused to form an input dataset for subsequent waveform generation. The feature vectors are normalized to ensure data consistency. Historical feature vector samples are obtained from the input dataset, and clustering algorithms are used to group different emotional state features. Monitoring parameters are then updated based on the grouping results to improve real-time performance.

[0008] Preferably, step S2 specifically includes: The biomarker features are input into the generator module of the generative adversarial network to generate an initial waveform sequence. The generator uses a convolutional layer to process the feature vector and output frequency and pulse width parameters. Noise is injected into the initial waveform sequence, and brainwave chaos is simulated through random noise distribution to obtain a variant waveform sequence, wherein the noise intensity is dynamically adjusted according to the user's historical response data; The discriminator module of the generative adversarial network is used to evaluate the similarity between the variant waveform sequence and the real brainwave. If the similarity is lower than a preset threshold, the generator parameters are iteratively optimized until convergence. Phase information is extracted from the variant waveform sequence and synchronized with the brainwave oscillation activity to generate the final stimulation waveform for current delivery. The synchronization matching adopts a phase-locked loop mechanism to ensure timing consistency. The variation history of the final stimulus waveform is recorded, and the noise algorithm parameters are updated by fusing feedback signals to maintain long-term variability and prevent tolerance accumulation.

[0009] Preferably, in step S3, the brain electric field model is generated through multi-physics coupling modeling, and the specific details of integrating skin impedance data and EEG feedback signals are as follows: Multiphysics coupling modeling integrates skin impedance and skull conductivity parameters, and uses the finite element method to divide the head mesh model and solve the current distribution equation to obtain the initial electric field model; Based on the initial electric field model, skin impedance data is input, and a microprobe array is used to measure impedance changes and update the model boundary conditions. If the impedance change exceeds a preset threshold, the current path is recalculated. Neural response indicators are obtained from EEG feedback signals, integrated into an electric field model to optimize distribution parameters, and generated an individualized brain electric field model for electrode configuration. The individualized brain electric field model outputs an electrode weight matrix that guides the controller to adjust the output intensity to compensate for uneven distribution.

[0010] Preferably, in step S3, a controller is used to dynamically adjust the electrode output to ensure that the current is delivered to the target brain region. The specific details are as follows: The controller receives skin impedance data and EEG feedback signals, calculates error signals, and fine-tunes the current intensity using a proportional-integral-differential algorithm. The error signal is based on the difference between the electric field intensity of the target brain region and the actual distribution. Based on the error signal, an adjustment command is generated and applied to the output weight of each electrode to ensure that the current loss is controlled within a preset range. The real-time feedback loop is extracted from the adjustment instructions, the controller parameters are updated to adapt to changes in physiological variables, and the adjustment log is recorded for model iteration. The brain region response during the current delivery process is monitored. If the response deviation exceeds a threshold, a secondary adjustment is triggered to maintain accuracy.

[0011] Preferably, the specific details of step S4, which involves processing EEG data based on a time-series analysis model to predict precursors of emotional lows and trigger intervention procedures, are as follows: The time series analysis model receives EEG data streams and uses a long short-term memory network to train historical sequences to identify emotional patterns, where the network input integrates biomarker features and timestamps. The prediction time window is calculated based on the emotion pattern. If a trough precursor is detected, an early warning signal is generated and a preheating intervention procedure is triggered. The prediction accuracy index is extracted from the early warning signal, and the network parameters are iteratively optimized through cloud computing power to shorten the response time. The preheating intervention procedure activates buffered stimulation, stabilizes the neural state based on the variant waveform sequence, and records the intervention results for subsequent prediction and calibration.

[0012] Preferably, the specific details of generating a sentiment map in step S4 to optimize subsequent solutions are as follows: The emotion map integrates EEG logs and conditioning records, and generates a two-dimensional map to represent the distribution of different emotional states through visualization algorithms; Based on the association of the two-dimensional mapping analysis mode, update the personalized solution parameters and upload them to the cloud platform; Optimization feedback is obtained from the cloud platform and applied to the waveform generation and current delivery modules to improve overall effectiveness; The sentiment map is periodically reviewed. If changes in the map indicate a decline in effectiveness, the threshold of the prediction model is adjusted and a new scheme is generated.

[0013] An intelligent transcranial microcurrent stimulation system for neuromodulation includes: The signal acquisition and processing module is used to acquire multi-channel EEG signals through a dry electrode array, perform filtering preprocessing, extract biomarker features based on Fourier transform, and construct a feature vector containing time-domain and frequency-domain indices. The waveform generation module is used to generate personalized microcurrent stimulation waveforms based on the biomarker features through a generative adversarial network, and to introduce noise into the waveforms to simulate brainwave chaos, thereby achieving phase synchronization with brainwave oscillations. The electric field modeling and control module is used to generate individualized brain electric field models based on multi-physics coupling modeling, integrate skin impedance data and EEG feedback signals, and dynamically adjust the electrode output weights through the controller to optimize the delivery of current to the target brain region. The prediction and intervention module is used to process EEG data based on time series analysis models, identify emotional patterns and predict the precursors of emotional lows, trigger pre-intervention procedures, and generate emotion maps to optimize subsequent treatment plans. The data management and optimization module is used to record EEG logs, treatment records, and adjustment history. It iteratively updates model parameters through a cloud platform to achieve adaptive optimization of the system.

[0014] Therefore, the present invention employs the above-mentioned intelligent transcranial microcurrent stimulation method and system for nerve modulation, which has the following beneficial effects: (1) The user's EEG signal is collected by a high-density dry electrode array, key biomarker features are extracted based on fast Fourier transform, and a dynamic variable microcurrent waveform that is phase-synchronized with the user's EEG oscillation activity is generated by using a generative adversarial network combined with a noise algorithm to avoid neural adaptive tolerance. (2) Multi-physics coupling modeling is used to integrate skin impedance and brain tissue characteristics. The current path is calculated by the finite element method. Combined with a proportional-integral-differential controller, the electrode output is dynamically adjusted to ensure that the current is accurately delivered to the target brain region. (3) Based on the prediction of the precursors of emotional low points by long short-term memory network, the system automatically triggers pre-warming intervention and generates an emotion map to support remote optimization schemes, thereby achieving personalized and precise emotion regulation and neural stimulation, significantly improving long-term effectiveness and user experience.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an embodiment of an intelligent transcranial microcurrent stimulation method for neural modulation according to the present invention; Figure 2 This is an architectural diagram of an embodiment of an intelligent transcranial microcurrent stimulation system for neural modulation according to the present invention; Figure 3 This is a schematic diagram showing the connection structure of various devices in an embodiment of an intelligent transcranial microcurrent stimulation system for neural modulation according to the present invention. Detailed Implementation

[0017] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0018] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] Example Please see Figure 1 This invention provides an intelligent transcranial microcurrent stimulation method for neural modulation, comprising: S1. Acquire electroencephalogram (EEG) signal data using a dry electrode array and extract biomarker features based on Fourier transform. This includes: A dry electrode array was used to acquire multi-channel EEG signal data, and signal preprocessing and filtering were applied during the acquisition process to remove noise interference and obtain a clean EEG signal sequence. The system inputs the clean EEG signal sequence into the Fourier transform module, calculates the power spectral density, and extracts the relative ratio of alpha and theta waves as a biomarker feature. If the relative ratio exceeds a preset threshold, it is determined to be an emotional fluctuation state and the corresponding timestamp is marked. Feature vectors are constructed based on biomarker features, and time-domain and frequency-domain indicators are fused to form an input dataset for subsequent waveform generation. The feature vectors are normalized to ensure data consistency. Historical feature vector samples are obtained from the input dataset, and clustering algorithms are used to group different emotional state features. Monitoring parameters are then updated based on the grouping results to improve real-time performance.

[0020] S2. Personalized microcurrent stimulation waveforms are generated based on biomarker characteristics. These waveforms dynamically vary to simulate brainwave chaos and avoid neural adaptive tolerance, with the waveforms being phase-synchronized with brain electrical oscillations. This includes: The biomarker features are input into the generator module of the generative adversarial network to generate an initial waveform sequence. The generator uses a convolutional layer to process the feature vector and output frequency and pulse width parameters. Noise is injected into the initial waveform sequence, and the chaos of brain waves is simulated by random noise distribution to obtain a variant waveform sequence. The noise intensity is dynamically adjusted according to the user's historical response data. The noise algorithm injects random noise into the waveform generation process to ensure the non-repetitiveness of each stimulus sequence, and adjusts the noise intensity according to the user's historical response data to improve long-term effectiveness.

[0021] The discriminator module of the generative adversarial network is used to evaluate the similarity between the mutated waveform sequence and the real brainwave. If the similarity is lower than a preset threshold, the generator parameters are iteratively optimized until convergence. The discriminator uses a fully connected layer to evaluate the degree of biological randomness matching.

[0022] Phase information is extracted from the variant waveform sequence and synchronized with the brainwave oscillation activity to generate the final stimulation waveform for current delivery. The synchronization matching adopts a phase-locked loop mechanism to ensure timing consistency. The variation history of the final stimulus waveform is recorded, and the noise algorithm parameters are updated by fusing feedback signals to maintain long-term variability and prevent tolerance accumulation.

[0023] S3. A brain electric field model is generated through multiphysics coupling modeling, integrating skin impedance data and EEG feedback signals. A controller dynamically adjusts the electrode output to deliver current to the target brain region. The specific details of generating the brain electric field model through multiphysics coupling modeling, integrating skin impedance data and EEG feedback signals, are as follows: Multiphysics coupling modeling integrates skin impedance and skull conductivity parameters, and uses the finite element method to divide the head mesh model and solve the current distribution equation to obtain the initial electric field model; Based on the initial electric field model, skin impedance data is input, and a microprobe array is used to measure impedance changes and update the model boundary conditions. If the impedance change exceeds a preset threshold, the current path is recalculated. Neural response indicators are obtained from EEG feedback signals, integrated into an electric field model to optimize distribution parameters, and generated an individualized brain electric field model for electrode configuration. The individualized brain electric field model outputs an electrode weight matrix that guides the controller to adjust the output intensity to compensate for uneven distribution.

[0024] The controller dynamically adjusts the electrode output to ensure that current is delivered to the target brain region. Details are as follows: The controller receives skin impedance data and EEG feedback signals, calculates error signals, and fine-tunes the current intensity using a proportional-integral-differential algorithm. The error signal is based on the difference between the electric field intensity of the target brain region and the actual distribution. Based on the error signal, an adjustment command is generated and applied to the output weight of each electrode to ensure that the current loss is controlled within a preset range. The real-time feedback loop is extracted from the adjustment instructions, the controller parameters are updated to adapt to changes in physiological variables, and the adjustment log is recorded for model iteration. The brain region response during the current delivery process is monitored. If the response deviation exceeds a threshold, a secondary adjustment is triggered to maintain accuracy.

[0025] S4. Based on time series analysis models, EEG data is processed to predict early signs of emotional lows and trigger intervention procedures. At the same time, an emotion map is generated to optimize subsequent treatment plans.

[0026] The specific details of using time-series analysis models to process EEG data, predict precursors to mood troughs, and trigger intervention procedures are as follows: The time series analysis model receives EEG data streams and uses a long short-term memory network to train historical sequences to identify emotional patterns, where the network input integrates biomarker features and timestamps. The prediction time window is calculated based on the emotion pattern. If a trough precursor is detected, an early warning signal is generated and a preheating intervention procedure is triggered. The prediction accuracy index is extracted from the early warning signal, and the network parameters are iteratively optimized through cloud computing power to shorten the response time. The preheating intervention procedure activates buffered stimulation, stabilizes the neural state based on the variant waveform sequence, and records the intervention results for subsequent prediction and calibration.

[0027] The specific details of generating a sentiment map to optimize subsequent solutions are as follows: The emotion map integrates EEG logs and conditioning records, and generates a two-dimensional map to represent the distribution of different emotional states through visualization algorithms; Based on the association of the two-dimensional mapping analysis mode, update the personalized solution parameters and upload them to the cloud platform; Optimization feedback is obtained from the cloud platform and applied to the waveform generation and current delivery modules to improve overall effectiveness; The sentiment map is periodically reviewed. If changes in the map indicate a decline in effectiveness, the threshold of the prediction model is adjusted and a new scheme is generated.

[0028] like Figure 2 Furthermore, an intelligent transcranial microcurrent stimulation system for neuromodulation is provided, comprising: The signal acquisition and processing module is used to acquire multi-channel EEG signals through a dry electrode array, perform filtering preprocessing, extract biomarker features based on Fourier transform, and construct a feature vector containing time-domain and frequency-domain indices. The waveform generation module is used to generate personalized microcurrent stimulation waveforms based on the biomarker features through a generative adversarial network, and to introduce noise into the waveforms to simulate brainwave chaos, thereby achieving phase synchronization with brainwave oscillations. The electric field modeling and control module is used to generate individualized brain electric field models based on multi-physics coupling modeling, integrate skin impedance data and EEG feedback signals, and dynamically adjust the electrode output weights through the controller to optimize the delivery of current to the target brain region. The prediction and intervention module is used to process EEG data based on time series analysis models, identify emotional patterns and predict the precursors of emotional lows, trigger pre-intervention procedures, and generate emotion maps to optimize subsequent treatment plans. The data management and optimization module is used to record EEG logs, treatment records, and adjustment history. It iteratively updates model parameters through a cloud platform to achieve adaptive optimization of the system.

[0029] Example 1 like Figure 3 It adopts a development framework based on the MVVM (Model-View-ViewModel) design pattern, and builds a complete closed-loop architecture with smart hardware as the core, integrating cloud computing and mobile terminal interaction. The core application carrier is the integrated headband device and its supporting mobile application. The deep collaboration between software and hardware realizes the full-process functions of EEG signal acquisition, analysis, intelligent intervention and data management. The overall technical system takes into account real-time performance, accuracy and medical-grade reliability.

[0030] The headband, as the core hardware terminal, is responsible for EEG signal acquisition, microcurrent stimulation delivery, and simple operation feedback. Its hardware components include the headband body, a sensing-stimulation electrode array, and a core control module. The headband is made of lightweight medical-grade materials and incorporates a high-density dry electrode array for EEG signal acquisition, along with a microprobe array for real-time skin impedance monitoring. The core control module consists of a high-performance MCU, a dedicated EEG signal processing chip, a high-precision digital-to-analog converter (DAC), and a Bluetooth communication module. It serves as the central hub for data acquisition, processing, and output control, providing the hardware foundation for subsequent signal analysis and intervention execution.

[0031] The ability of microprobes to simultaneously perform both impedance measurement (signal acquisition) and discharge (electrostimulation) functions lies in the coordinated design of their hardware materials and circuit timing. At the hardware level, their surfaces are modified with nanomaterials such as platinum black or conductive polymers to create a microstructure that combines low impedance (facilitating sensitive recording of weak signals) with high charge injection capability (safely releasing current through reversible electrochemical reactions, avoiding hydrolysis). In terms of operating mechanism, a time-division multiplexing strategy controlled by a phase-locked loop clock is employed, enabling high-speed switching between "recording mode" and "stimulation mode" on microsecond or even shorter timescales. Specifically, the system inserts an extremely short time window within each stimulation pulse cycle to switch to measurement mode, tracking instantaneous phase changes in the EEG signal to ensure that each output of the stimulation waveform is synchronized with the brain's endogenous rhythms. Meanwhile, this high-speed switching mechanism enables the system to acquire updated skin impedance and neural response data in real time during the intervals of microcurrent stimulation, providing immediate feedback parameters to the PID controller. This ensures that the next stimulation pulse can be dynamically calibrated based on the latest physiological state, achieving precise closed-loop regulation. Through this design, the microprobe alternately completes sensing and discharging tasks on the same physical channel, macroscopically realizing a bidirectional neural interface function that integrates reading and writing.

[0032] The mobile application serves as the core user interaction platform, integrating key modules such as emotion map visualization, treatment data recording, early warning notifications, and personalized plan adjustments. The emotion map visually displays the user's treatment progress and emotional trends using dynamic charts, and combines AI-generated stimulation strategy suggestions to optimize long-term treatment plans. Simultaneously, the application supports users actively recording emotion logs, which are deeply integrated with passively monitored EEG data from the device to further improve the accuracy of the emotion prediction model, building an efficient interactive bridge between the user and the system.

[0033] The system software adopts a layered architecture, divided into an embedded firmware layer, a core algorithm engine, and an application and cloud layer. Each layer performs its specific function and works collaboratively. The embedded firmware layer runs a real-time operating system (RTOS), primarily responsible for hardware driver management and basic signal preprocessing, ensuring stable operation of hardware devices and initial processing of raw signals. The core algorithm engine is the core of the system's intelligence, integrating an LSTM prediction model, a GAN waveform generation model, and a PID closed-loop control algorithm, supporting key functions such as emotion prediction, personalized stimulus waveform generation, and precise current regulation. The cloud-based GAN waveform generation model is trained and updated, while the local MCU only runs a lightweight generator (inference). The application and cloud layer provides a user-friendly interface, secure data storage services, and convenient remote update capabilities, enabling edge-cloud data interoperability and optimizing the user experience.

[0034] At the communication mechanism level, the system constructs a multi-dimensional, highly reliable communication architecture, balancing real-time interaction between devices and data synchronization between the device and the cloud. The device and mobile terminal use Bluetooth Low Energy (BLE) 5.2 as the primary wireless communication method. Leveraging its low power consumption, high stability, low latency, and high throughput, it adapts to the needs of portable medical devices worn for extended periods, maintaining a stable connection even in mixed usage modes. The EEG signals collected by the headband are pre-processed and transmitted to the mobile app via Bluetooth with AES-256 end-to-end encryption, complying with medical-grade data privacy protection standards. The app acts as an intermediary, providing an intuitive user interface and synchronizing encrypted data to the cloud server. The cloud employs a distributed computing architecture, supporting the storage and in-depth analysis of massive amounts of user data, while also providing over-the-air (OTA) firmware upgrade functionality to ensure the device is always in optimal operating condition. To address network anomalies, the system incorporates multi-layered fault tolerance mechanisms, including local caching of unuploaded data, automatic reconnection, and breakpoint resumption, comprehensively ensuring data integrity and system reliability. In addition, the internal EEG sensor array and the core control module communicate at high speed via I2C or SPI protocols to ensure real-time acquisition and low-latency transmission of EEG data. To enable cross-platform interaction, the system has built a cloud interface based on RESTful API. The device can upload encrypted EEG data to the cloud through this interface and call the LSTM prediction model to complete the identification and early warning of mood troughs.

[0035] Signal processing and feature extraction are the foundation for the system to achieve accurate emotion monitoring and intelligent intervention. The entire process balances signal purity and feature effectiveness. In the signal preprocessing stage, algorithms such as bandpass filtering, notch filtering, and independent component analysis (ICA) are used to effectively remove noise such as power frequency interference and motion artifacts, making the EEG signal purer. After preprocessing, the signal is used for feature extraction through Fast Fourier Transform (FFT) and wavelet transform to accurately identify biomarkers closely related to emotional states, such as the power spectral density of alpha waves and theta waves and their ratio, providing core feature data for subsequent model analysis and intervention decisions.

[0036] The system's core intelligent intervention function relies on the synergistic operation of an LSTM prediction model, a GAN waveform generation model, and a PID closed-loop control algorithm. This overcomes the limitations of traditional fixed intervention modes, enabling proactive, personalized, and precise intervention. The LSTM (Long Short-Term Memory) network, as the core algorithm for predictive intervention, uses deep learning and modeling of EEG time-series data to accurately predict emotional lows more than 20 minutes in advance, allowing ample time for proactive intervention. The GAN (Generative Adversarial Network) overcomes the limitations of traditional fixed waveforms by dynamically generating non-repetitive microcurrent pulse sequences based on extracted user EEG features. It also incorporates a noise algorithm simulating biological randomness to effectively avoid neural adaptation tolerance issues, achieving personalized stimulus waveform generation. The PID closed-loop control algorithm, as the core regulation mechanism for intervention execution, integrates skin impedance, neural feedback signals, and individual physiological difference data to dynamically adjust the output weights and current parameters of each electrode, overcoming current attenuation problems caused by individual physiological differences and ensuring intervention effectiveness. Furthermore, the AI ​​model is deployed on the MCU using lightweight frameworks such as TensorFlow Lite, reducing reliance on external computing resources and ensuring real-time processing capabilities on the device.

[0037] The high-precision hardware configuration provides solid technical support for signal acquisition, processing, and intervention execution, with all parameters meeting the requirements for medical-grade real-time monitoring and precise control. During EEG signal transmission, the analog-to-digital converter supports 24-bit high resolution and a sampling rate of up to 1000Hz, fully capturing subtle changes in EEG signals to meet real-time monitoring needs. A high-precision digital-to-analog converter (DAC) paired with a voltage-controlled constant current source circuit ensures that the output current error is controlled within 0.1 microamps. Each electrode point is equipped with an independent voltage-controlled constant current source, providing hardware assurance for precise current delivery. In the closed-loop control stage, the PID algorithm achieves millisecond-level adjustment speed, further improving the real-time performance and accuracy of the intervention.

[0038] The system's processing is based on neural oscillation synchronization theory and nonlinear dynamics. The core of microcurrent stimulation lies in achieving phase synchronization with the neural activity of the target brain region, and this synchronization depends on real-time capture and rapid response of EEG characteristics. To ensure processing effectiveness, the system rigorously optimizes key parameters, including waveform frequency range, pulse width modulation accuracy, current intensity resolution, and closed-loop feedback delay time. Clear processing quality assessment standards have also been established, requiring EEG synchronization of over 95%, adjustment delay of less than 50 milliseconds, and emotion prediction accuracy of no less than 85%. Its long-term stability and reliability have been verified through rigorous testing.

[0039] In the signal transmission and current delivery stages, a high-precision, adaptive processing pipeline was constructed, balancing signal acquisition quality with intervention execution accuracy. The sensing-stimulation electrode array within the headband uses built-in microprobes to achieve real-time skin impedance monitoring. The acquired EEG signals are amplified, converted from analog to digital, and then transmitted to the core control module. The entire process relies on a high-precision signal processing pipeline composed of algorithms such as bandpass filtering, notch filtering, and ICA to effectively remove artifacts and noise interference, ensuring signal purity and reliability. For current delivery, a PID closed-loop control algorithm is used to dynamically adjust the output voltage based on skin impedance monitoring data. An independent voltage-controlled constant current source for each electrode further compensates for current attenuation caused by individual physiological differences, ensuring the accuracy and effectiveness of microcurrent stimulation.

[0040] The system's implementation follows a standardized process, seamlessly connecting each stage from initialization to passive monitoring and active intervention. Data is encrypted throughout the process to ensure operational standardization, data security, and intervention accuracy. First, in the initialization phase, a comprehensive hardware self-check and skin impedance calibration are performed to confirm all electrode connections are secure and establish initial EEG baseline data for the user, providing a reference benchmark for subsequent monitoring and intervention. Next, the system enters passive monitoring mode. EEG signals, after bandpass filtering and ICA algorithm preprocessing, are segmented into fixed time windows and input into an LSTM model for real-time analysis, continuously monitoring changes in the user's emotional state. When the LSTM model detects precursors to an emotional low, the system automatically triggers an active intervention mechanism. The PID controller dynamically adjusts the current output parameters, while a GAN model is invoked to generate a personalized microcurrent waveform sequence based on the user's individual characteristics, implementing precise microcurrent stimulation. Throughout the process, all EEG data, intervention parameters, emotional state information, etc., are encrypted using AES-256 and synchronized via Bluetooth to a mobile app and cloud server for subsequent in-depth analysis, data tracing, and emotional map visualization. The core code mainly covers three modules: signal acquisition and filtering logic on the embedded side, data parsing and model calling process of cloud API interface, and dynamic rendering function of emotion map on mobile app. The code of each module works together to support the implementation of the entire system function.

[0041] Therefore, the present invention employs the above-mentioned intelligent transcranial microcurrent stimulation method and system for neuromodulation to achieve personalized and precise emotion regulation and neurostimulation, significantly improving long-term effectiveness and user experience.

[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent transcranial microcurrent stimulation system for neural modulation, characterized in that, include: The signal acquisition and processing module is used to acquire multi-channel EEG signals through a dry electrode array, perform filtering preprocessing, extract biomarker features based on Fourier transform, and construct a feature vector containing time-domain and frequency-domain indices. The waveform generation module is used to generate personalized microcurrent stimulation waveforms based on the biomarker features using a generative adversarial network, and to introduce noise into the waveforms to simulate brainwave chaos, thereby achieving phase synchronization with brainwave oscillations. Specific features include: The biomarker features are input into the generator module of the generative adversarial network to generate an initial waveform sequence. The generator uses a convolutional layer to process the feature vector and output frequency and pulse width parameters. Noise is injected into the initial waveform sequence, and brainwave chaos is simulated through random noise distribution to obtain a variant waveform sequence, wherein the noise intensity is dynamically adjusted according to the user's historical response data; The discriminator module of the generative adversarial network is used to evaluate the similarity between the variant waveform sequence and the real brainwave. If the similarity is lower than a preset threshold, the generator parameters are iteratively optimized until convergence. Phase information is extracted from the variant waveform sequence and synchronized with the brainwave oscillation activity to generate the final stimulation waveform for current delivery. The synchronization matching adopts a phase-locked loop mechanism to ensure timing consistency. The variation history of the final stimulus waveform is recorded, and the noise algorithm parameters are updated by fusing feedback signals to maintain long-term variability and prevent tolerance accumulation. The electric field modeling and control module is used to generate individualized brain electric field models based on multi-physics coupling modeling, integrate skin impedance data and EEG feedback signals, and dynamically adjust the electrode output weights through the controller to optimize the delivery of current to the target brain region. The prediction and intervention module is used to process EEG data based on time series analysis models, identify emotional patterns and predict the precursors of emotional lows, trigger pre-intervention procedures, and generate emotion maps to optimize the treatment plan. The data management and optimization module is used to record EEG logs, treatment records, and adjustment history. It iteratively updates model parameters through a cloud platform to achieve adaptive optimization of the system.

2. The intelligent transcranial microcurrent stimulation system for neural modulation according to claim 1, characterized in that, The process involves acquiring multi-channel EEG signals via a dry electrode array, performing filtering preprocessing, and extracting Fourier transform-based biomarker features to construct a feature vector containing time-domain and frequency-domain indices. A dry electrode array acquires multi-channel EEG signal data, and during the acquisition process, signal preprocessing and filtering are applied to remove noise interference to obtain a clean EEG signal sequence. The clean EEG signal sequence is input into a Fourier transform module to calculate the power spectral density and extract the relative ratio of alpha waves and theta waves as biomarker features. If the relative ratio exceeds a preset threshold, it is determined to be an emotional fluctuation state and the corresponding timestamp is marked. Feature vectors are constructed based on biomarker features, and time-domain and frequency-domain indicators are fused to form an input dataset for waveform generation. The feature vectors are normalized to ensure data consistency. Historical feature vector samples are obtained from the input dataset, and clustering algorithms are used to group different emotional state features. The parameters are then updated based on the grouping results.

3. The intelligent transcranial microcurrent stimulation system for neural modulation according to claim 1, characterized in that, The specific details of generating a brain electric field model based on multiphysics coupling modeling, integrating skin impedance data and EEG feedback signals, are as follows: Multiphysics coupling modeling integrates skin impedance and skull conductivity parameters, and uses the finite element method to divide the head mesh model and solve the current distribution equation to obtain the initial electric field model; Based on the initial electric field model, skin impedance data is input, and a microprobe array is used to measure impedance changes and update the model boundary conditions. If the impedance change exceeds a preset threshold, the current path is recalculated. Neural response indicators are obtained from EEG feedback signals, integrated into an electric field model to optimize distribution parameters, and generated an individualized brain electric field model for electrode configuration. The individualized brain electric field model outputs an electrode weight matrix that guides the controller to adjust the output intensity to compensate for uneven distribution.

4. The intelligent transcranial microcurrent stimulation system for neural modulation according to claim 1, characterized in that, The method of dynamically adjusting the electrode output weights through a controller to optimize current delivery to the target brain region includes: The controller receives skin impedance data and EEG feedback signals, calculates error signals, and fine-tunes the current intensity using a proportional-integral-differential algorithm. The error signal is based on the difference between the electric field intensity of the target brain region and the actual distribution. An adjustment command is generated based on the error signal and applied to the output weight of each electrode to control the current loss within a preset range. The real-time feedback loop is extracted from the adjustment instructions, the controller parameters are updated to adapt to changes in physiological variables, and the adjustment log is recorded for model iteration. The brain region response during the current delivery process is monitored, and if the response deviation exceeds a threshold, a secondary adjustment is triggered.

5. The intelligent transcranial microcurrent stimulation system for neural modulation according to claim 1, characterized in that, The process of processing EEG data based on a time-series analysis model to identify emotional patterns and predict precursors to emotional lows, triggering a pre-warming intervention procedure, includes: The time series analysis model receives EEG data streams and uses a long short-term memory network to train historical sequences to identify emotional patterns, where the network input integrates biomarker features and timestamps. The prediction time window is calculated based on the emotion pattern. If a trough precursor is detected, an early warning signal is generated and a preheating intervention procedure is triggered. The prediction accuracy index is extracted from the early warning signal, and the network parameters are iteratively optimized through cloud computing power to shorten the response time. The preheating intervention procedure activates buffered stimulation, stabilizes neural states based on variant waveform sequences, and records intervention results for predictive calibration.

6. The intelligent transcranial microcurrent stimulation system for neural modulation according to claim 1, characterized in that, The specific details of generating a sentiment map to optimize subsequent solutions are as follows: The emotion map integrates EEG logs and conditioning records, and generates a two-dimensional map to represent the distribution of different emotional states through visualization algorithms; Based on the association of the two-dimensional mapping analysis mode, update the personalized solution parameters and upload them to the cloud platform; Optimization feedback is obtained from the cloud platform and applied to the waveform generation and current delivery modules; The sentiment map is periodically reviewed. If changes in the map indicate a decline in effectiveness, the threshold of the prediction model is adjusted and a new scheme is generated.

Images