Transcranial alternating current stimulation method and system with automatic adjustment function

Abstract

The present application provides a transcranial alternating current stimulation method and system with automatic adjustment function, which relates to the field of transcranial alternating current stimulation technology. The method includes acquiring and compressing sampling multi-channel EEG signals, performing multi-channel reconstruction using physiological priors and correlation information of adjacent brain regions, extracting target brain wave characteristic parameters, and outputting a two-stage transcranial alternating current signal of pre-stimulation and post-stimulation based on these parameters; by introducing an adjacency matrix or graph Laplacian operator in the compressed sensing acquisition and reconstruction link, high-fidelity recovery of cross-brain region oscillations can be maintained under low sampling rate and noise conditions; combining general and individual priors, hierarchical screening and dynamic fine-tuning of characteristic frequency bands of different brain regions are performed; in the pre-stimulation stage, the target brain wave phase is first locked, and in the post-stimulation stage, the stimulation intensity of each brain region is independently or collaboratively adjusted to meet the needs of multi-brain region coupling regulation, thereby improving the closed-loop intervention effect on neural states such as insomnia and anxiety.

Description

Technical Field

[0001] The present application relates to the technical field of transcranial alternating current stimulation, and in particular to a transcranial alternating current stimulation method and system with automatic adjustment function. Background Art

[0002] With the continuous development of non-invasive neuromodulation technologies, transcranial alternating current stimulation (tACS) has attracted increasing attention as an aid in improving abnormal brain function, such as insomnia and anxiety. Existing studies have shown that electrical stimulation applied at the appropriate frequency and timing can interact with the brain's endogenous rhythms, thereby affecting the activity level or phase characteristics of brain regions. However, typical tACS protocols are often based on single-channel or limited-channel EEG acquisition and output stimulation at a fixed frequency or phase in an open-loop mode. This lacks adaptability to real-time EEG changes and the coordinated state of multiple brain regions. Furthermore, to acquire sufficient EEG signals in high-noise environments, high sampling rates and long processing delays are often required, increasing system power consumption and data bandwidth pressure. When synchronization or coupling anomalies exist across multiple brain regions, independent channel processing and stimulation alone cannot capture the overall abnormal fluctuations across brain regions in a timely manner, nor can they properly allocate stimulation frequency, intensity, and timing to each region. In practical applications, these limitations often lead to inadequate extraction of key EEG features and inflexible and inaccurate closed-loop feedback, compromising the effectiveness of transcranial alternating current stimulation in conditions such as insomnia and anxiety. Summary of the Invention

[0003] In response to the deficiencies of the existing technology, the present application provides a transcranial alternating current stimulation method and system with automatic adjustment function.

[0004] In a first aspect, the present application provides a transcranial alternating current stimulation method with an automatic adjustment function, comprising:

[0005] Acquire multi-channel EEG signals from multiple brain regions of the target user, and perform compressed sensing sampling on the multi-channel EEG signals based on the characteristic frequency bands of each brain region to generate compressed data;

[0006] Performing multi-channel EEG reconstruction on the compressed data, and extracting target EEG characteristic parameters corresponding to the target state in combination with physiological priors of each brain region;

[0007] Based on the target brain wave characteristic parameters, determining the transcranial alternating current stimulation output parameters for each target brain area and outputting the signal;

[0008] dividing the transcranial alternating current stimulation output parameters into prior stimulation parameters and subsequent stimulation parameters;

[0009] The prior stimulation parameters are used to pre-stimulate the target brain area and determine the preliminary phase information of the brain waves of the target brain area; the subsequent stimulation parameters are dynamically adjusted based on the preliminary phase information.

[0010] As an optional implementation, performing multi-channel EEG reconstruction on the compressed data includes:

[0011] Based on the placement of scalp electrodes or the functional division of brain regions, an adjacency matrix is ​​constructed to represent the correlation between channels in adjacent brain regions.

[0012] The penalty term corresponding to the adjacency matrix is ​​introduced into the reconstruction optimization objective, so that the penalty value is increased when the phase or amplitude difference of the reconstructed signal of each pair of adjacent channels exceeds a preset range, and the penalty value is included in the objective function together with the data consistency term and the sparse regularization term for solution.

[0013] As an optional implementation, when performing the multi-channel EEG reconstruction, a reconstruction difference between any two channels marked as adjacent according to the adjacency matrix is ​​detected;

[0014] When the reconstruction difference exceeds a preset reconstruction difference threshold, the measurement matrix is ​​adjusted to obtain new compressed data in a subsequent iteration, and the multi-channel EEG reconstruction is continued based on the new compressed data.

[0015] As an optional implementation, when performing multi-channel EEG reconstruction on the compressed data, the method includes:

[0016] Calculating a graph Laplacian based on the adjacency matrix, and incorporating the graph Laplacian into a reconstruction optimization objective;

[0017] During the iterative solution process, the graph Laplacian operator is used to constrain the differences between the channels so that the reconstructed signals of adjacent channels remain locally consistent in phase or amplitude.

[0018] As an optional implementation, when performing graph Laplacian-constrained multi-channel EEG reconstruction, local differences between channels marked as adjacent to the adjacency matrix are detected;

[0019] In response to determining that the local difference corresponds to a true brain region feature rather than noise interference, reducing the constraint weight of the graph Laplacian operator to allow the corresponding channel to maintain a reconstruction result different from that of the adjacent channel;

[0020] In response to determining that the local difference belongs to noise interference, the constraint weight of the graph Laplacian operator is maintained or increased to reduce the reconstruction deviation between the interfered channel and the adjacent channel.

[0021] As an optional implementation, the step of extracting target brain wave characteristic parameters corresponding to the target state by combining physiological priors of each brain region includes:

[0022] Based on the preset brain region function priors, corresponding target frequency bands are set for the frontal lobe, parietal lobe, and occipital lobe respectively;

[0023] After the multi-channel EEG reconstruction is completed, a first round of frequency band screening is performed on the reconstructed signal of each brain region according to the functional prior;

[0024] Combined with the user's individualized priors established in baseline acquisition or historical records, the user's typical peak frequency or phase is located within the target frequency band obtained in the first round of screening;

[0025] In response to the difference between the individualized prior and the functional prior, the frequency band interval of the corresponding brain region is locally expanded.

[0026] As an optional implementation, in response to detecting a strong peak or phase that is not covered by the individualized prior and the functional prior, marking the corresponding frequency band as a suspicious interval and increasing the sampling density or iteration weight in the next reconstruction cycle;

[0027] In response to repeated detection of a strong peak or a phase that persists and is independent of noise interference, the peak range corresponding to the corresponding brain region in the individualized prior is updated.

[0028] As an optional implementation, when target brain wave abnormalities are detected in multiple brain regions and the association weights of any two brain regions in the adjacency matrix are greater than a preset threshold, a coupling abnormality brain region group is generated;

[0029] According to the target brain wave characteristics of each brain region in the abnormally coupled brain region group, respectively determining and storing the corresponding initial parameters of the prior stimulation, including stimulation frequency, initial phase and pre-stimulation duration;

[0030] Simultaneously, applying a prior stimulation to the abnormally coupled brain region group and generating a phase-locking index, and if the phase-locking indexes are all greater than or equal to a first threshold after the prior stimulation ends, outputting identification information indicating that a subsequent stimulation phase has begun;

[0031] In the post-stimulation stage, personalized stimulation intensity or frequency parameters for each brain region are calculated and stored, and closed-loop stimulation output is performed on each brain region based on the frequency parameters.

[0032] As an optional embodiment, in the post-stimulation stage, the phase difference or energy difference of each brain region in the abnormally coupled brain region group is acquired in real time, and a coupling degree measurement result is generated and updated;

[0033] When the target brain wave amplitude difference or phase deviation of any brain region is detected to be lower than a second threshold, the stimulation intensity or frequency of the brain region is downgraded based on the current coupling measurement result, and the downgrade parameters are stored for subsequent iterations;

[0034] When it is monitored that the phase or energy shift of the abnormally coupled brain region group occurs again as a whole, the downshift parameter and the coupling degree measurement result are used to determine whether to regenerate the initial parameters of the previous stimulation.

[0035] In a second aspect, the present application provides a transcranial alternating current stimulation system with automatic adjustment function, comprising:

[0036] An acquisition module is used to obtain multi-channel EEG signals from multiple brain regions of the target user, and perform compressed sensing sampling on the multi-channel EEG signals based on the characteristic frequency bands of each brain region to generate compressed data;

[0037] A processing module, configured to perform multi-channel EEG reconstruction on the compressed data, and extract target EEG characteristic parameters corresponding to the target state in combination with physiological priors of each brain region;

[0038] an output module, which determines the transcranial alternating current stimulation output parameters for each target brain region based on the target brain wave characteristic parameters and outputs the signal;

[0039] The output module is further used to divide the transcranial alternating current stimulation output parameters into prior stimulation parameters and subsequent stimulation parameters;

[0040] The prior stimulation parameters are used to pre-stimulate the target brain area and determine the preliminary phase information of the brain waves of the target brain area; the subsequent stimulation parameters are dynamically adjusted based on the preliminary phase information.

[0041] Compared with the existing technology, the present application proposes an EEG reconstruction method that takes into account the fusion of multi-channel compressed sensing sampling and physiological priors, and models the correlation between adjacent brain regions based on the adjacency matrix. On the one hand, by adding sparse regularization or graph Laplacian operators in a low sampling rate environment, stable EEG reconstruction under high noise is achieved; on the other hand, by combining general functional priors with individualized priors to perform hierarchical screening of the target frequency band of each brain region, the peak and phase of abnormal brain waves can be accurately extracted. On this basis, the use of a two-stage output scheme of prior stimulation and post-stimulation helps to lock the phase of the target brain region first, and then strengthen or maintain individual brain regions. Especially when there are coupling fluctuations between multiple brain regions, the overall synchronization of adjacent brain regions can be enhanced through unified initial stimulation and dynamic adjustment methods. Compared with the existing technology, the present application also reduces sampling and computational overhead, improves the flexibility of reconstruction and closed-loop control, and adapts to the needs of multi-brain region coordinated control in scenarios such as insomnia and anxiety. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A flow chart of a transcranial alternating current stimulation method with automatic adjustment function provided in an embodiment of the present application;

[0043] Figure 2 A schematic diagram of a priori information provided in an embodiment of the present application;

[0044] Figure 3 A flowchart of constructing an adjacency matrix provided in the embodiment of this application

[0045] Figure 4 Schematic diagram of a transcranial alternating current stimulation system with automatic adjustment function provided in an embodiment of the present application.

[0046] Reference numerals: 10, acquisition module; 20, processing module; 30, output module. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0048] See also Figure 1 FIG. 1 is a flow chart of a transcranial alternating current stimulation method with an automatic adjustment function provided in an embodiment of the present application. The transcranial alternating current stimulation method with an automatic adjustment function includes S101 to S104, wherein:

[0049] S101: Acquire multi-channel EEG signals distributed across multiple brain regions of a target user, and perform compressed sensing sampling on the multi-channel EEG signals based on characteristic frequency bands of each brain region to generate compressed data;

[0050] S102: Perform multi-channel EEG reconstruction on the compressed data, and extract target EEG characteristic parameters corresponding to the target state in combination with physiological priors of each brain region;

[0051] S103: Based on the target brain wave characteristic parameters, determining the transcranial alternating current stimulation output parameters for each target brain area, and outputting the signals;

[0052] S104: Divide the transcranial alternating current stimulation output parameters into prior stimulation parameters and subsequent stimulation parameters; wherein, the prior stimulation parameters are used to pre-stimulate the target brain area and determine the preliminary phase information of the brain waves of the target brain area; the subsequent stimulation parameters are dynamically adjusted based on the preliminary phase information.

[0053] In a specific implementation, the present application arranges electrodes in multiple target brain areas of the user's head and forms a multi-channel EEG acquisition pathway.

[0054] Specifically, in order to reduce hardware resource usage and transmission bandwidth, a measurement matrix can be designed based on the target brain wave frequency bands commonly found in each brain region. The coefficients of the measurement matrix can be implemented in hardware circuits or software, thereby performing a linear projection operation on the original EEG signal in the analog or digitization stage and outputting compressed data. By making the measurement matrix sparse or structured, the main information related to the target brain region frequency band can be obtained at a lower sampling rate, thereby increasing sensitivity to key oscillation components. If an increase in noise or a decrease in reconstruction quality is detected at a certain stage, it is possible to temporarily switch to a spare measurement matrix or increase the sampling frequency to ensure the accuracy of subsequent reconstruction.

[0055] For example, the ADS1299 acquisition chip can be used in conjunction with an STM32 series microcontroller or field-programmable gate array (FPGA) to perform linear projection of the measurement matrix and data compression. Electroencephalogram (EEG) signals are preprocessed using fixed-coefficient sparse filtering units in a Verilog or C language environment. The compressed data is then transmitted to a host computer or cloud using Wi-Fi or Bluetooth. Sparse reconstruction algorithms such as orthogonal matching pursuit (OMP) and iterative hard thresholding (IHT) are then used in Python or the Matrix Lab (MATLAB) environment to restore key oscillation frequency band information. When noise levels increase or reconstruction quality degrades, firmware can dynamically switch to an alternate measurement matrix or increase the sampling rate to effectively control bandwidth and power consumption while maintaining critical signal quality.

[0056] After obtaining the compressed data, the present application transmits it to the processing unit to perform multi-channel EEG reconstruction. The reconstruction process can be based on cross-channel covariance analysis or an optimization model based on sparse priors to distinguish between effective brain waves and noise components. In order to make the system stable in a noisy environment, a robust estimation submodule for Gaussian noise, electromyographic interference or sporadic pulse noise can be introduced in the reconstruction, and stable brain wave components can be extracted through sparse regularization or low-rank decomposition. In the reconstructed multi-channel EEG waveform, the physiological prior of each brain region can be used to strengthen the focus on specific frequency bands. For example, when monitoring insomniac users, focus on retaining slow wave components in the range of 0.5 to 1 Hz, and when monitoring anxious users, focus on the energy and phase of alpha waves in the range of 8 to 12 Hz.

[0057] In the process of extracting target brain wave characteristic parameters described in this application, the typical peak frequency refers to the frequency corresponding to the main power peak presented by the reconstructed signal of a certain brain area within the target frequency band. Specifically, after performing spectral analysis on multi-channel EEG, the power spectrum or amplitude distribution can be calculated within a specified frequency band (such as 0.5-1Hz, 8-12Hz, etc.), and the frequency point with the highest energy is searched as the "typical peak frequency" of the brain area in the current state. This means that if the maximum power point of a brain area in the range of 8-12Hz appears at 10Hz, 10Hz can be regarded as the typical peak frequency of the brain area.

[0058] During specific implementation, the system will combine the user's individualized prior data, such as the actual peak distribution of the user's alpha wave measured during the baseline phase. If multiple observations show that its maximum power is mostly concentrated in the range of 9.3Hz to 9.7Hz, the algorithm will regard this frequency band as the user's typical peak range and focus on the power characteristics within this interval during this reconstruction analysis. If the actual detected power peak deviates slightly from the previous range, the system can determine whether it is necessary to expand the search bandwidth or perform a higher-resolution reconstruction to confirm whether the deviation is a short-term noise or a new typical peak. In this way, the "typical peak frequency" is used in the present invention to quantitatively describe the main energy concentration characteristics of each brain region in the target frequency band, and provide an accurate phase and frequency matching reference for subsequent prior / later stimulation.

[0059] It should be noted that the target state is not limited to anxiety, and the solution disclosed in this application is also applicable to insomnia, depression, etc.

[0060] When extracting the target brainwave characteristic parameters, the energy, phase, and peak frequency of the brainwave in the target frequency band can be calculated for each brain region. If the difference between the individualized baseline level and the current measurement exceeds a pre-set threshold, the corresponding brain region can be marked as requiring key regulation and given a higher priority in the subsequent stimulation output link. For the marked brain regions, the system will generate transcranial alternating current stimulation output parameters, including stimulation frequency, phase, and intensity.

[0061] For example, after completing the reconstruction of the multi-channel EEG and obtaining the brainwaves in the target frequency band, the energy, phase, and peak frequency of each brain region can be calculated one by one through Python or MATLAB, and these parameters can be compared with the individual baseline levels. If the difference in a certain brain region exceeds the set threshold, the system will mark it as requiring key control. In the subsequent control phase, combining the existing ADS1299 and FPGA architecture, the system will generate the corresponding transcranial alternating current stimulation (tACS) output parameters based on the marked brain regions, including stimulation frequency, phase, and intensity, to prioritize these key brain regions and help correct or strengthen the required brainwave oscillations.

[0062] To avoid potential discomfort or loss of phase synchronization caused by large-scale stimulation during initial control, pre-stimulation parameters can be applied first, for a short duration and at a relatively low current amplitude to observe brain responses. If complete phase information is extracted during this process and the user does not experience excessive discomfort, the system can automatically enter the post-stimulation phase, appropriately increasing or adjusting the stimulation intensity and frequency to synchronize with the real-time EEG phase.

[0063] In practice, the system first enters the pre-stimulation phase: for example, to the pre-marked frontal or parietal lobes, a tACS signal of approximately 0.8 mA is applied, with the frequency set based on the previously measured alpha wave peak (e.g., 10 Hz). At this point, the FPGA, in conjunction with the microcontroller, acquires real-time EEG and evaluates the phase response of the target brain region through phase detection (implemented using a Hilbert transform). During the 5-second pre-stimulation phase, if both EEG phase stability (e.g., phase lock value (PLV) greater than 0.6) and user comfort (e.g., subjective assessment score less than or equal to 2) meet thresholds, the system determines that stimulation parameters can be further increased. Next, the system automatically enters the post-stimulation phase, strengthening the regulation effect by increasing the tACS intensity from 0.8 mA to 1.5 mA or fine-tuning the frequency (such as searching for the optimal synchronization frequency within the range of 9.5 to 10.5 Hz). If the phase synchronization detected by the EEG decreases significantly (PLV is less than 0.4) or the user experiences obvious discomfort (subjective assessment score greater than or equal to 3), the system will revert to the previous stimulation parameters or stop stimulation to ensure safety and acceptability.

[0064] In addition, a time-slice-based scheduling mechanism can be established within the processing unit to determine the end time of the prior stimulation and the start time of the subsequent stimulation by polling the EEG reconstruction results of each brain region. If the processing unit detects that the prior stimulation cannot successfully obtain phase synchronization or the user experiences a high-intensity discomfort reaction, it will automatically reduce the stimulation amplitude or suspend the pre-stimulation, and monitor the noise level and EEG characteristics of the brain region again. If it is detected in the post-stimulation stage that the target waveform of the brain region has been stably enhanced, the maintenance time of the current parameters can be extended or the stimulation intensity can be phased down. Conversely, if the brain wave drops or phase drifts during the post-stimulation stage, the processing unit will dynamically adjust the stimulation output parameters to restore the output phase alignment and ensure that the power does not exceed the safety limit. The above process is continuously repeated in the system, thereby achieving automated closed-loop regulation for insomniac and anxious users while meeting safety constraints.

[0065] In some wearable implementations, the acquisition and reconstruction functions can be integrated on the same hardware platform, and the reconstructed key characteristic parameters can be sent to peripheral devices or the cloud for further analysis via a low-power wireless module. The switching timing between pre-stimulation and post-stimulation can be fine-tuned according to historical statistical laws to take into account the personalized needs of different users. For users with a significantly low slow wave or alpha wave base, the duration of the pre-stimulation can be extended to ensure sufficient initial phase lock. For sensitive people, the current amplitude and increment rate can be strictly limited during the pre-stimulation phase.

[0066] This differentiated strategy enables targeted management of sleep and mood in both clinical and home settings, while mitigating potential adverse reactions during stimulation. This approach, with clear hardware and algorithm support at each step, is well-suited to the low-power and high-precision requirements of real-world applications.

[0067] For example, four electrodes can be placed on the frontal and parietal lobes of the user's head, forming a total eight-channel EEG acquisition array. The hardware can utilize an analog front-end module with a multi-channel programmable gain amplifier, which bandpass filters the collected EEG signals to a range of approximately 0.1Hz to 40Hz. The analog output of each channel is then fed into an analog-to-digital converter (ADC) via a microcontroller-controlled multiplexer. The conversion accuracy can be selected to be 12-bit or 16-bit, depending on the specific power budget. To reduce overall data transmission volume and achieve low-power operation, a sparse or structured measurement matrix can be pre-loaded on the embedded system. The sampled EEG signals are linearly projected in a random or structured sparse manner to obtain compressed data with lower dimensionality. For insomnia intervention, the frontal and parietal lobes are often associated with slow-wave sleep, so the measurement matrix can focus on preserving as much frequency domain information as possible in the 0.5Hz to 1Hz range.

[0068] In the software part, to speed up the processing, the compressed sensing reconstruction algorithm can be run on a digital signal processor with floating-point computing capabilities or a low-power FPGA. The reconstruction process can be divided into two main stages: after the processor is started, the prior of the cross-channel covariance matrix is ​​loaded, and the optimization equation is constructed in combination with the sparse assumption of the slow wave component, and then the EEG data of each channel is restored through iterative solution. After completing the multi-channel reconstruction, the software estimates the power density and phase in the range of 0.5Hz to 1Hz based on the timing waveform of each channel, and compares it with the user's baseline data to determine the current deep sleep tendency.

[0069] After deriving the characteristic parameters, the system calculates the corresponding transcranial alternating current stimulation output parameters for each brain region, including the stimulation frequency and phase alignment method. Initially, the output parameters are set to the pre-stimulation mode, and an AC current amplitude of 0.3mA to 0.5mA and a frequency range close to the slow wave can be selected. The pre-stimulation time can be set from about tens of seconds to two minutes, depending on the user's tolerance to stimulation and EEG response speed. If the system detects that the phase of the slow wave signal is basically locked with the external stimulation during this period and the user does not report obvious discomfort, the software will automatically switch the stimulation of the brain region to the post-stimulation mode, and the amplitude can be slightly increased to 0.6mA to 0.8mA, or more flexible phase fine-tuning can be performed within the same frequency range. If the slow wave intensity is detected to have increased significantly and remained stable, the system will extend the post-stimulation interval or reduce the amplitude to save energy and prevent overstimulation.

[0070] Furthermore, in conventional multi-channel reconstruction, the system often regards each channel as independent of each other or constrains it only through a global sparse prior. In this way, when adjacent electrode areas are perturbed by similar noise, local reconstruction instability or phase incoherence across brain regions is likely to occur.

[0071] To this end, as an optional implementation, spatial or functional correlation modeling of adjacent brain regions can be introduced into the reconstruction algorithm, thereby mathematically achieving synchronous correction or joint sparse expression of adjacent signals.

[0072] In a specific implementation, in the multi-channel EEG acquisition link, an adjacency matrix can be constructed based on the layout of the electrodes on the scalp to represent the spatial or functional adjacency between the channels. The adjacency matrix can be obtained based on the geometric distance of the scalp electrodes or the functional division of brain regions known in relevant references. For electrode pairs that are close to each other or functionally coupled, the corresponding elements in the adjacency matrix can be set to a larger weight, otherwise a smaller or zero weight. When performing multi-channel EEG reconstruction, a regularization term based on the adjacency matrix can be added on the basis of the original sparse regularization or low-rank constraint. Exemplarily, the following form can be used:

[0073] ;

[0074] in, Represents the EEG signal representation or coefficient vector to be solved (can also be regarded as a matrix depending on the situation). When a specific transform domain (such as wavelet domain, discrete cosine domain or discrete Fourier domain) is used in the sampling and reconstruction process, Considered as a set of coefficients in the transform domain; if processed directly in the time domain, It is regarded as a sequence of discrete sampling points of a multi-channel EEG signal. Represents the measurement matrix, which is used to implement compressed sensing sampling. At the hardware or algorithm level, Projecting the raw EEG signal into a lower-dimensional measurement space. This is usually achieved using random sparse or structured sparse designs and can be implemented in embedded hardware as sparse multiplication. Represents the compressed measurement data (observation data) obtained by actual sampling. Due to the use of compressed sensing method, The dimension will be lower than that of the original EEG signal of complete sampling, but a relatively high-fidelity EEG waveform can be restored in the reconstruction.

[0075] It is called data fidelity term or data consistency term, which is used to measure the current solution In the measurement space and the actual observation The deviation between the two. Commonly used two-norm square ( ) is used to measure the deviation. If the item is smaller, it means that through When mapped back to the measurement space, the closer it is to the true sample value.

[0076] and are weight coefficients, which are used to adjust the importance of different regularization terms in the optimization objective. A larger value means that the algorithm pays more attention to The a priori hypothesis represented by The larger the value, the greater the joint constraint on adjacent brain regions. Will be stronger.

[0077] Used to express conventional prior constraints on EEG signals or their coefficients, such as:

[0078] Sparse priors (e.g. norm, total variation, etc.), used to encourage signals or coefficients to maintain sparsity in a certain domain, thereby suppressing noise; low-rank prior, if the multi-channel EEG is regarded as a matrix, its rank can be measured by the nuclear norm or other low-rank norms, encouraging the existence of correlation structures between different leads; other common denoising or smoothing constraints, such as Intensity suppression under the norm or threshold processing based on the wavelet domain.

[0079] This part is used to suppress noise overall and emphasize the compressibility or smoothness of the signal in a given domain.

[0080] represents a regularization term related to the correlation between adjacent brain regions, and A is an adjacency matrix (or weighted adjacency matrix) that characterizes the adjacent or functional relationships between channels. The construction and form of the adjacency matrix A can be found at the end of this article and will not be detailed here.

[0081] There are many ways to design the regularization term:

[0082] As an optional implementation, adjacent smoothness constraints can be used to impose certain smoothness or phase consistency requirements on the reconstruction results of adjacent brain region channels. For example, a penalty term is added when the phase or amplitude difference between adjacent channels at the same time is too large.

[0083] As an optional implementation, a graph Laplacian penalty can be used to regard the EEG signal as a distribution on a graph structure, and to encourage a certain smoothness or low-frequency characteristics between adjacent nodes (channels) through graph signal processing methods (such as the graph Laplacian operator).

[0084] As an optional implementation, sparse coupling can be used. When it is known that adjacent brain regions have synchronous or coherent oscillations in pathological states of insomnia or anxiety, the two can be encouraged to present synchronous non-zero coefficients at the same frequency to avoid meaningless independent activation or noise spikes.

[0085] Through this constraint, the solution of each channel can be adaptively adjusted during optimization to reduce local unreasonable oscillations or noise interference, while better preserving the synchronous rhythms that may exist across brain regions and improving the depiction of real physiological activities.

[0086] This means finding the optimal X that minimizes the objective function. This problem is typically solved using iterative methods such as the split-Bregman method, the alternating direction method of multipliers (ADMM), and the iterative soft thresholding method (ISTA). The key is selecting an appropriate algorithm that maintains low computational overhead while converging to a high-fidelity solution under sparse priors and adjacency constraints.

[0087] The above formula aims to balance data fidelity with multiple a priori constraints, enabling high-quality reconstruction of multi-channel EEG signals in real-world environments with low sampling rates (downsampling due to compressed sensing) and noise interference, thereby solving the following technical problems:

[0088] Cross-brain waveform segmentation: If adjacent brain regions should theoretically have good synchronization or similarity, but due to noise or insufficient sampling, they show extreme differences after reconstruction, adjacency constraints can be used to maintain reasonable similarity between them.

[0089] Local noise: Adjacent brain regions are often subject to similar environmental interference. By combining constraints, the noise can be prevented from being excessively amplified in a particular lead.

[0090] Pathological resonance detection: If users with insomnia or anxiety have common characteristic fluctuations in certain adjacent brain regions, this formula can help highlight such common patterns in reconstruction and more accurately identify brainwave characteristics related to pathological conditions.

[0091] In specific implementations, penalties can be set for the phase or amplitude differences between adjacent channels in the frequency domain to reduce unreasonable cross-region mutations and ensure that the phases of slow waves or alpha waves in adjacent brain regions are smoother.

[0092] As an optional implementation, a graph-based signal processing approach can be adopted, treating each channel as a graph node and constraining cross-channel smoothness or local consistency through the graph Laplacian operator defined by the adjacency matrix, thereby preserving coherent oscillation characteristics across brain regions under low sampling rate conditions.

[0093] If it is known that two brain regions often exhibit coupled oscillations in insomnia or anxiety scenarios, the corresponding matrix weights can be increased so that the reconstruction algorithm pays more attention to maintaining synchronization or common sparsity between the two regions.

[0094] This contiguity constraint allows for better extraction of EEG rhythms that are conducted or coupled across brain regions during reconstruction, reducing reconstruction distortion caused by single-point interference or local noise. When the system detects excessive differences in reconstructions between adjacent brain regions, it can trigger additional robustness correction steps within the algorithm, such as temporarily increasing the sampling frequency, using a denser measurement matrix, or employing a stricter threshold suppression method, thereby enhancing resistance to environmental interference.

[0095] For example, if, when monitoring the border area between the forehead and the parietal lobe, it is found that the slow wave phase or alpha wave phase of these two adjacent channels has abnormal jumps, the adjacency constraint regularization term can force the algorithm to bring the solutions of the two channels closer during the calculation, ensuring that the waveforms across regions are more consistent. Conversely, if there is indeed a pathological condition that causes an abnormal phase difference between the forehead and the parietal lobe, the system can compare the adjacency matrix and the global prior to determine whether the difference really exists. If it is confirmed that the difference is real brain activity rather than noise, the algorithm can distinguish between real signals and interference with the support of the remaining channels, thereby improving the accuracy of identifying abnormal pathological conditions.

[0096] When setting prior or subsequent stimulation parameters for different brain regions in subsequent transcranial alternating current stimulation control, if the processing unit detects significant frequency band power drops in multiple adjacent brain regions during the reconstruction phase through adjacency constraints, these regions can be simultaneously designated as high-priority stimulation targets to synchronously enhance their brain waves, avoiding the overall effect of stimulating only a single brain region. For example, in an anxious state, if both the frontal lobe and the anterior parietal lobe are found to exhibit low alpha wave energy, and both have a large weight in the adjacency matrix, phase-locked multi-channel stimulation can be applied to these two regions to restore synchronization between their alpha waves and jointly alleviate anxiety symptoms.

[0097] Through the above methods, the new multi-channel EEG reconstruction method increases the utilization of information connections between adjacent brain regions in the original scheme, solving technical problems that may be faced by single-channel or independent channel methods, such as waveform fragmentation across brain regions, insufficient robustness in local noise processing, and difficulty in maintaining phase consistency between adjacent brain regions. This improvement can complement the aforementioned prior / post stimulation mechanism:

[0098] The multi-channel EEG reconstruction stage uses proximity constraints to provide more accurate spatial distribution and phase estimation;

[0099] During the stimulation phase, pre-stimulation is used to lock the phase first, and then post-stimulation is used to consolidate or enhance the synchronization of brain areas, thereby more accurately regulating brain wave oscillations in insomnia or anxiety scenarios.

[0100] By introducing correlation modeling between adjacent brain regions during the reconstruction phase, this application can further enhance the fidelity and robustness of compressed sensing reconstruction, provide a more reliable basis for brainwave characteristics in subsequent partitioned stimulation strategies, and significantly improve the accuracy of cross-brain fluctuation detection. This not only technically addresses the vulnerability of single-channel or independent-channel methods to complex noise interference, but also provides more comprehensive support for synchronous closed-loop regulation of multiple functionally related brain regions.

[0101] As an optional implementation, the EEG channels distributed in multiple brain regions can be regarded as several graph nodes, and the neighboring nodes of each node under the scalp electrode position or brain region functional division are determined, and an adjacency matrix is ​​constructed based on this. The adjacency matrix is ​​used to calculate the graph Laplacian operator to smooth or coherently constrain the cross-brain region distribution in the multi-channel EEG reconstruction process. After the compressed data obtained in the sampling stage is transmitted to the processing unit, the processing unit can combine the graph Laplacian operator with sparse regularization or low-rank prior during the iterative solution, on the one hand to ensure the restoration of key brain waves at low sampling rates, and on the other hand to reduce unreasonable waveform mutations or phase jumps between adjacent brain regions with the help of graph smoothing.

[0102] During the initialization phase, the user's electrode layout information or clinical data can be read first, channel nodes can be assigned to each brain region, and edge weights can be determined between adjacent nodes. If two brain regions are close in location or have a high degree of synchronization during baseline monitoring, they can be given a larger weight, thereby forming a closer connection relationship in the graph structure. After obtaining the graph Laplacian operator through the adjacency matrix, when the processing unit performs multi-channel reconstruction, each round of iteration will add additional smoothing constraints based on the graph structure to suppress the amplitude or phase differences of adjacent nodes in the same frequency band. When a channel is detected to have a significant deviation and a large inconsistency with the adjacent channel, the system can correct the reconstruction result of the channel based on the gradient information of the Laplacian term, thereby maintaining the possibility that local mutations reflect the real signal while minimizing outliers caused by noise or artifacts.

[0103] In practical applications, a smoothing coefficient corresponding to the graph Laplacian operator can be set to adjust the trade-off between the degree of smoothing of adjacent nodes and the global sparsity prior. If the reconstruction error is detected to be too high or the coupling between adjacent brain regions is insufficient, the system can appropriately increase the smoothing coefficient to enhance the maintenance of cross-brain coherence. If the user has clear pathological features that cause a truly abnormal waveform in a certain brain region, the influence of adjacent smoothing on the channel can be temporarily weakened after the clinical characteristics or prior annotation of the brain region are retrieved to avoid confusing the true abnormality with noise. This can achieve a balance between global smoothing across brain regions and local specific changes.

[0104] When the reconstructed multi-channel signal better preserves the coherent oscillations of adjacent brain regions, the subsequent pre-stimulation and post-stimulation stages can be regulated according to more accurate brain wave phase information. If slow wave enhancement is detected in both the frontal and parietal lobes, similar stimulation phases or frequencies can be set for these two brain regions during closed-loop control. If it is detected that the phase of a single brain region is significantly different from that of the surrounding brain regions and cannot be reconciled after multiple iterative reconstructions, the system can consider increasing the sampling frequency or replacing the measurement matrix to confirm whether the difference is noise or an actual local abnormality. During use, if insufficient bandwidth or computing resources are detected, the graph smoothing coefficient or the update frequency of some channels can be temporarily reduced to maintain the real-time performance of the overall algorithm.

[0105] This graph signal processing approach can enhance the accuracy of identifying synchronized activity across brain regions at low sampling rates, making it particularly suitable for the multi-brain region coupled oscillations commonly seen in people with insomnia and anxiety. Compared to existing simple penalties based on adjacency matrices, the use of the graph Laplacian operator allows for more flexible modeling of brain network structures while being algorithmically compatible with existing sparse or low-rank reconstruction modules. Combining a two-stage current output method of pre-stimulation and post-stimulation, along with multi-channel EEG reconstruction results based on graph smoothing, allows for more accurate synchronous phase locking or phase adjustment in subsequent closed-loop control, providing an alternative implementation approach for different clinical or home scenarios.

[0106] For example, several electrodes can be placed in the frontal, parietal, and occipital lobes of the user's head, forming a twelve-channel EEG acquisition array. The acquisition hardware can combine an ADS1299 analog front end with an FPGA for linear projection, while an STM32 series microcontroller controls data acquisition and wireless transmission. Initially, the system reads the scalp electrode locations and functional partitioning information, creates a corresponding graph node for each channel, and calculates edge weights between adjacent nodes based on the geometric distance between electrodes, functional correlations reported in the literature, or phase synchronization measured in pre-experiments. Electrode pairs that exhibit high phase coherence during baseline monitoring (e.g., coupling between the frontal and parietal lobes in the slow-wave frequency band) are assigned a higher weight in the adjacency matrix. However, if two channels are located far apart and lack synchronization characteristics based on preliminary testing, the corresponding position in the adjacency matrix is ​​assigned a lower weight.

[0107] For example, if it is monitored that the leads in the frontal, parietal and occipital regions simultaneously show a significant decrease in slow wave energy, then in the subsequent closed-loop stimulation, the system can mark these three channels together as high-priority objects to ensure that a closer phase-locked frequency or a higher stimulation intensity is allocated during transcranial alternating current stimulation. If it is detected that one of the channels continues to differ greatly from the surrounding channels, but the difference cannot be narrowed by graph smoothing, the processing unit will perform supplementary sampling on the channel to determine whether it is indeed a real brain abnormality. If it is confirmed to be noise interference, the system will suppress the interference within the neighborhood under the constraint of the graph Laplacian operator, and will not affect the overall phase judgment; if it is confirmed to be a pathological event, different prior or subsequent stimulation parameters will be set for the channel separately in the subsequent stimulation output, so as to take into account both cross-brain region coupling and local personalized regulation.

[0108] For example, the fast iterative update of the graph Laplace regularization term can be achieved through fixed-point operations on the FPGA or DSP side, or offline verification can be performed using high-precision floating-point operations in Python or MATLAB environments. The smoothing coefficient can be dynamically adjusted between 0.01 and 0.1 according to the patient's condition or noise level: when the system detects that adjacent brain areas should maintain synchronization but show random deviations, the smoothing coefficient is appropriately increased; when there is a clear lesion area and the difference really needs to be retained, the smoothing coefficient is reduced or the adjacent constraint on the lesion area is stopped. The multi-channel EEG signal obtained on this basis can better reflect the correlation between the real brain areas in terms of phase and amplitude, laying a more reliable data foundation for the accurate phase locking and intensity distribution of the preceding and subsequent stimulations.

[0109] In this way, by incorporating the spatial or functional correlations between adjacent brain regions into the graph Laplace constraints, the system can still maintain the overall coherence of cross-brain region oscillations in low sampling rates and high noise environments, and provide more flexible multi-brain region coordinated regulation for conditions such as insomnia or anxiety in subsequent closed-loop transcranial alternating current stimulation.

[0110] As an optional implementation, after multi-channel EEG reconstruction is complete, the primary frequency bands of regions such as the frontal, parietal, or occipital lobes can be limited to corresponding intervals based on the common functional properties of these brain regions. For example, in the frontal lobe, the slow wave range of 0.5-1Hz is prioritized, while in the parietal or occipital lobes, the characteristic frequency bands of alpha or theta waves are emphasized. This universal functional prior can quickly filter out relatively more noteworthy intervals from the full frequency band in a low-sampling environment, reducing redundant calculations of irrelevant frequencies.

[0111] After screening the target interval, the specific peak frequency and phase can be fine-tuned for each brain region in combination with the user's individual prior information. If the true peak of a certain brain region is detected to deviate from the default prior, such as the common range of alpha waves is 8-12Hz but the user's actual peak is mostly concentrated around 12.5Hz, then in subsequent analysis and stimulation decisions, the attention bandwidth will be appropriately expanded to 12-13Hz to avoid ignoring the key waveform caused by individual deviation. If this deviation only occurs in a certain measurement, the system can perform more sample acquisition in the next timing window to verify whether there is a stable energy peak or phase synchronization feature to prevent transient interference from being mistakenly included in the prior model.

[0112] When the general functional prior conflicts with the individualized prior or fails to cover a new characteristic peak, the system can temporarily mark the suspected frequency band of the brain region as a "suspicious segment" and increase the sampling density or iteratively reconstruct the weights in the next stage. If multiple verifications show that the new peak has a clear and lasting energy advantage, the processing unit will update the baseline prior for the brain region so that it will always take this new peak range into account in future analyses of the same user. In this way, even if abnormal oscillations or unique individual characteristics appear outside the common frequency band, they will not be ignored by the traditional fixed prior model.

[0113] Through this hierarchical prior fusion approach, each brain region is first screened across a wide frequency range based on its functional attributes. This is then combined with the individualized data recorded from the user's baseline for localized fine-tuning. This ensures consistency with the universal functional division of brain regions while capturing the user's differentiated activity patterns. The extracted target brainwave features are therefore closer to the true physiological state, more accurately reflecting the actual performance of individual brain regions in terms of energy, phase, and peak frequency, providing a more reliable basis for the subsequent calculation of parameters for pre- and post-stimulation.

[0114] If the application detects the sudden emergence of a new frequency band in a brain region that is inconsistent with existing prior information and exhibits high amplitude or significant phase synchronization, the system identifies this as a potential marker of pathology or unique brainwave activity. Subsequently, the system increases the compressed sensing sampling frequency or switches to a measurement matrix more appropriate for this characteristic to quickly determine whether closed-loop control of this brain region, distinct from conventional frequency bands, is necessary. By layering and integrating common functions with individual differences, the entire feature extraction process achieves both stability and flexibility, enabling the full extraction of brainwave information relevant to the target state even in low-sampling-rate, noisy environments.

[0115] For example, see Figure 2 , Figure 2 A schematic diagram of a priori information provided for an embodiment of the present application; wherein, the system will store two types of prior data for each brain region during initialization. The first type is functional prior, which is used to describe the common main frequency bands and waveform characteristics of the brain region, such as the preset slow wave interval of the frontal lobe is 0.5-1Hz, and the alpha wave interval of the occipital lobe or parietal lobe is 8-12Hz. The second type is individualized prior, which is used to record the actual peak frequency and time domain characteristics measured by the user during baseline acquisition or multiple previous uses. Baseline acquisition can be completed through continuous monitoring of multi-channel EEG signals, usually lasting several hours and covering different states such as quiet rest or light sleep, so as to obtain the typical brain wave distribution of each brain region of the user.

[0116] During specific analysis, after multi-channel reconstruction, the system performs a functional prior match on the reconstructed signals of each brain region. The software module prioritizes searching for amplitude and phase peaks within the 0.5-1 Hz range within the spectrum of the frontal lobe signal. If a valid peak or phase synchronization is found, it is preliminarily labeled as the primary slow wave activity in that brain region. Subsequently, based on the individualized prior, if the user's past slow wave peaks were more concentrated around 0.7 Hz, the search will be prioritized within the 0.6-0.8 Hz range within the 0.5-1 Hz subregion. The actual peak frequency detected is then compared with the individual baseline. If the difference is within 0.05 Hz, the algorithm determines that it matches the user's prior and records the corresponding phase and energy as the core features of this monitoring session. If the difference exceeds this range and the peak persists, the frequency range is marked as a "suspected new peak" and a flag is set to prompt the user to increase the sampling density or lower the threshold in subsequent iterations or the next measurement to determine whether the user's prior should be updated.

[0117] In this embodiment, if a user's frontal lobe signal essentially matches the functional prior (0.5-1Hz), but repeatedly exhibits a distinct and stable peak at 1.05Hz during repeated sampling, the system will temporarily include the 1.0-1.1Hz interval in the key search range for the next round of reconstruction. After determining that activity is still present, the system will automatically extend the upper limit of the frontal lobe peak frequency in the user's baseline data from 1.0Hz to 1.1Hz to ensure that this newly emerged peak is not missed in future analyses. If this peak only appears briefly in a single measurement and does not reappear in subsequent time series windows, the system will not make any lasting changes to the individualized prior and will treat it as a short-term fluctuation or noise interference.

[0118] Through this two-level prior fusion process, the system not only ensures the rapid positioning of the functional characteristics of brain regions, but also performs step-by-step confirmation in combination with user historical records when peaks or phases that deviate from the normal distribution are found. If such deviations are determined to be true features, the baseline prior for subsequent analysis is updated; if they are determined to be noise or occasional fluctuations, the original prior is maintained. Technically, the entire implementation process relies on the hierarchical management of the prior library and the secondary calibration mechanism of local frequency band search. Focus strategies are set for the characteristic frequency bands of different brain regions, so that the results of each EEG reconstruction can accurately extract the truly valuable target brain wave parameters with a smaller amount of sampling and calculation. When combined with the setting of the prior stimulation and post-stimulation parameters, these updated peaks and phases will directly determine the corresponding phase-locked frequency and signal output amplitude, thereby achieving more personalized closed-loop adjustment.

[0119] Furthermore, when imbalances or common abnormalities exist in multiple adjacent brain regions at the same time, if the method of independently calculating stimulation parameters for a single brain region is still used, it may be impossible to coordinate the phase matching or intensity distribution between different brain regions, resulting in limited overall stimulation effects.

[0120] As an optional implementation, when the present application detects that there is a high correlation between adjacent brain regions and both exhibit abnormal target brain wave characteristics, these brain regions are regarded as a group of abnormally coupled brain regions, and combined with the spatial or functional relationship reflected by the adjacency matrix, unified timing and parameter management is performed on them during the stimulation stage.

[0121] Specifically, before pre-stimulation, similar or identical initial stimulation parameters can be set for these brain regions based on the dominant frequency band and phase differences of each channel within the group of abnormally coupled brain regions, so that they receive pre-stimulation within a similar frequency range or at a similar phase starting point. If it is monitored that some brain regions within the group have a high degree of phase locking, while others have not yet reached the threshold, the overall pre-stimulation duration can be extended until all highly correlated brain regions achieve phase alignment or the phase locking indicators of most channels meet the preset standards, and then the post-stimulation phase can be entered.

[0122] In this way, we no longer examine whether a single brain area completes pre-stimulation separately, but instead regard adjacent brain areas as a whole to evaluate whether the prior stimulation has achieved preliminary phase synchronization. This is particularly suitable for the joint regulation of the frontal lobe and parietal lobe or the parietal lobe and occipital lobe when slow waves or alpha waves are abnormal.

[0123] In the subsequent post-stimulation, the processing unit can further use the weights of the adjacency matrix to determine whether it is necessary to maintain synchronous stimulation of multiple brain regions. If it is detected that the group of brain regions with abnormal coupling has established good phase alignment for the previous stage, then in the subsequent stage, stimulation signals with the same frequency and minimal phase difference can be applied to these brain regions to consolidate the synchronous oscillation across brain regions and avoid each brain region entering a different frequency band or phase alone, thereby weakening the overall coupling. If one of the brain regions has recovered to a state close to the baseline, while other brain regions still have significant abnormalities, the stimulation intensity can be maintained or increased only for the latter, reducing repeated stimulation of the brain regions that have improved. If signs of overstimulation are detected in a certain brain region or the user's comfort level decreases, the post-stimulation output of this brain region can be reduced accordingly, and temporary attention can be paid to the abnormalities in the remaining neighboring regions.

[0124] Through such a dynamic allocation mechanism within the group, while meeting the synchronization needs of the group, it can also take into account the individual response of a single brain area, avoiding phase disorders caused by excessive or repeated stimulation.

[0125] In a specific implementation, the processing unit needs to perform timed or event-driven switching between the preceding stimulation and the following stimulation.

[0126] Among them, timed switching means that within the initially set prior stimulation time window, if the abnormally coupled brain area group has not yet achieved common phase alignment, the prior stimulation will continue to be maintained; if synchronization is still not achieved after the time expires, the system will enter a protection state and prompt the user or system administrator to avoid user discomfort under long-term high-intensity pre-stimulation.

[0127] Event-driven switching emphasizes real-time monitoring. Once the phase coherence of a group of abnormally coupled brain regions exceeds a preset threshold, post-stimulation is immediately initiated, with individual or small-scale adjustments made based on the remaining differences between each region. If new interference or phase shifts occur during post-stimulation, the processing unit can identify synchronization mismatches between the affected region and its neighbors based on the adjacency matrix. If necessary, it briefly reverts to the pre-stimulation mode to re-lock the common phase before switching back to post-stimulation.

[0128] By also referencing the adjacency matrix during the stimulation phase to coordinate the prior and subsequent regulation of multiple brain regions, the intra-group synergy can be better exerted when there are abnormal coupling fluctuations in adjacent brain regions, avoiding phase disconnection or intensity dispersion caused by independent stimulation of a single brain region.

[0129] For the multi-brain region abnormalities commonly observed in insomnia or anxiety scenarios, such as insufficient slow wave regulation between the frontal and parietal lobes or weakened alpha wave synchronization between the occipital and parietal lobes, the use of this coupled stimulation strategy can help to restore the coordinated oscillation of the brain region network on a large scale. In addition, when the abnormal amplitude or phase offset of a certain brain region is much larger than that of other brain regions, the system can stimulate it more strongly or more intensively in the later stage, while still maintaining the overall phase consistency with other adjacent brain regions, and will not destroy the integrity of the group by strengthening individual brain regions. In this way, the synchronous pre-locking of the prior stimulation is combined with the flexible individual adjustment of the subsequent stimulation, so that the present application can further play its guiding role in the cross-brain region correlation in the stimulation stage through the adjacency matrix in addition to the reconstruction stage, thereby providing a better closed-loop stimulation strategy for the case of multi-brain region coupling abnormalities.

[0130] For example, if there are adjacent electrodes in the frontal and parietal regions of a user's head, and the processing unit determines, based on the adjacency matrix, that these two regions exhibit high coupling during baseline monitoring, then the system can label these two brain regions as a group of abnormally coupled regions, and perform coordinated pre- and post-stimulation when determining the stimulation output.

[0131] Specifically:

[0132] During the pre-stimulation phase, the processing unit sets a similar initial frequency (e.g., 10 Hz) and a moderate current (e.g., 0.5 mA) for the forehead and parietal lobes, initiates the pre-stimulation process, and observes the degree of phase lock between the two brain regions. If one brain region detects a faster phase lock while the other still shows no clear signs of synchronization, the system automatically extends the duration of the pre-stimulation or slightly adjusts the phase offset of the slower-locking region to maximize phase alignment between the two regions within the pre-stimulation period. Only when the processing unit observes a phase difference between the two regions that is less than a preset threshold (e.g., 5 degrees) and the user's subjective comfort level is within an acceptable range does the pre-stimulation phase conclude and enter the post-stimulation phase.

[0133] During the post-stimulation phase, when initial phase synchronization has been achieved between the frontal and parietal lobes, the processing unit can further increase the stimulation intensity to 0.8-1.0 mA, maintaining phase-aligned stimulation at approximately 10 Hz to consolidate coupled oscillations. If stable synchronization is detected between the two regions and the EEG signal approaches baseline levels, the stimulation intensity of one region can be gradually reduced, thereby reducing unnecessary power consumption and user burden. If the other region has not yet reached the target, the system will extend the post-stimulation duration for that region or fine-tune the frequency again to 9.8-10.2 Hz to match the optimal phase window for that region. The system will not force the other region that has recovered to receive the same intensity. During this process, if the phase of one region is detected to be suddenly unlocked or the user experiences discomfort, the processing unit can temporarily reduce the intensity and observe whether it is necessary to return to a short period of prior stimulation to re-align the two regions.

[0134] This way, by knowing through the adjacency matrix that the frontal and parietal lobes have a high degree of cross-brain correlation, we can maximize synchronized pre-stimulation in the early stages, and then more flexibly reinforce or maintain individual stimulation in the later stages. Compared to stimulating the frontal and parietal lobes independently, this coupled pre-stimulation method effectively reduces the initial phase-locking time and allows for targeted treatment of suboptimal or deviated brain regions during subsequent post-stimulation, thereby more effectively enhancing the user's alpha wave activity and helping to alleviate anxiety or improve attention deficits.

[0135] For example, Table 1 gives an example adjacency matrix for representing the adjacent relationship and weights between the four brain region channels. Assume that four electrodes are laid on the scalp, named C1, C2, P1, and P2 (just as an example), and each element in the matrix represents a channel. With channel The degree of correlation between the two. A larger value indicates a closer relationship in terms of geometric position or function, while a zero value indicates no direct correlation. This matrix is ​​usually a symmetric matrix with zeros on the main diagonal.

[0136] Table 1 Adjacency matrix example table

[0137]

[0138] In this example:

[0139] The correlation between C1 and C2 is 0.65, which means that these two channels (probably located in adjacent frontal or central areas) show high phase coupling in functional or baseline tests;

[0140] The correlation between C1 and P1 is only 0.20, indicating that they are far away or their functional correlation is not significant;

[0141] The correlation between P1 and P2 reached 0.70, suggesting that a clear synchronization relationship can be observed between these channels (which may all be located near the parietal or occipital lobes); if the other positions are zero, it means that there is almost no direct coupling between the two or no strong correlation has been seen in the monitoring data.

[0142] When this adjacency matrix is ​​referenced in multi-channel EEG reconstruction or in the pre- / post-stimulation stage, the algorithm can determine which channels require synchronous pre-stimulation or subsequent enhanced regulation based on these correlations. It can also refer to the reconstruction results of its adjacent channels to constrain or correct the solution of a channel after detecting that a channel is interfered by noise, thereby maintaining overall consistency across brain regions.

[0143] In specific implementations, a variety of information and steps can be used to construct an adjacency matrix to quantify the correlation between brain region channels.

[0144] For example, see Figure 3 , Figure 3 A flowchart for constructing an adjacency matrix is ​​provided in an embodiment of the present application, wherein:

[0145] The first step is to collect the location information of the scalp electrodes:

[0146] Record the electrode layout coordinates on the user's head in a database, for example, using the international 10-20 system or measuring the distances between electrodes and anatomical landmarks at custom locations. If the international 10-20 system is used, the geometric distances between electrodes can be approximated by mapping simplified scalp coordinates to a reference template. Using high-precision positioning tools or 3D scanning can more accurately measure the coordinate differences between electrodes.

[0147] The International 10–20 System is an internationally recognized method for standardizing the placement of electroencephalogram (EEG) electrodes. This system ensures consistent and reproducible EEG recordings across individuals and laboratories, facilitating data comparison, analysis, and diagnosis.

[0148] The second step is to calculate the geometric distance weight:

[0149] Based on the electrode position coordinates from the previous step, the distance between each pair of electrodes is obtained using the Euclidean distance or other distance measurement. To facilitate subsequent fusion, this can be converted into a normalized weight between 0 and 1. For example, the distance d can be calculated by taking the inverse or exponential decay:

[0150] ;

[0151] in, represents the distance between the i-th electrode and the j-th electrode, A factor to adjust distance sensitivity. Reflects the geometric proximity between electrode i and electrode j. The closer the value is to 1, the shorter the physical distance between the electrodes is; a value close to 0 indicates that the distance between the two electrodes is far or there is almost no spatial correlation.

[0152] The third step is to obtain functional or physiological baseline data:

[0153] During the initial baseline test, the user undergoes multi-channel EEG monitoring for a specific duration (e.g., several minutes to several hours). The user can be placed in rest, eyes closed / open, light sleep, or other typical states to observe the degree of coupling, phase synchronization, or mutual information between pairs of electrodes in key frequency bands (e.g., alpha waves and slow waves).

[0154] The fourth step is to extract the functional association weight:

[0155] According to the EEG phase synchronization indicators (such as coherence coefficient, PLV, mutual information, Granger causality, etc.) or energy correlation in the baseline data, a functional coupling degree is assigned to each pair of electrodes in the range of 0 to 1, which is recorded as If the two electrodes continuously show high synchronization or energy correlation in the same target frequency band, then The higher the value, the lower it is.

[0156] Step 5: Merge distance and function information:

[0157] Combine the geometric distance weight and functional coupling degree in a certain form to obtain the final adjacency matrix weight .

[0158] For example, weighted average:

[0159] ;

[0160] in Used to control the relative importance of functional coupling and geometric distance;

[0161] Another example, product fusion:

[0162] ;

[0163] If any of them is small, the final weight will also decrease;

[0164] Another example is other custom functions: based on actual experience or experimental results, different forms of nonlinear combinations can be given to distance and functional coupling.

[0165] Step 6: Threshold and symmetry processing:

[0166] To ensure that the matrix is ​​easy to use and meets the algorithm requirements, the generated adjacency matrix is ​​processed as follows:

[0167] For example, symmetrization:

[0168] Let A(i,j)=A(j,i). If it is not naturally symmetric, take the mean, such as:

[0169] ;

[0170] in, is the adjacency matrix after symmetry processing.

[0171] For example, threshold cutoff:

[0172] If some weights are less than a preset threshold (such as 0.05 or 0.1), they are considered to have no obvious correlation and set to 0 to reduce noisy edges or unnecessary weak connections.

[0173] For example, normalization:

[0174] All elements of the matrix can be normalized to [0,1] to facilitate unified processing by subsequent algorithms.

[0175] After completing the above construction process, an adjacency matrix A that conforms to the actual electrode distribution and physiological coupling characteristics can be obtained, providing a basis for multi-channel EEG reconstruction, pre / post stimulation parameter scheduling, and cross-brain region coupling regulation.

[0176] Based on the same inventive concept, the embodiments of the present application also provide a transcranial alternating current stimulation system with automatic adjustment function corresponding to the transcranial alternating current stimulation method with automatic adjustment function. Since the principle of solving the problem by the system in the embodiments of the present application is similar to the above-mentioned transcranial alternating current stimulation method with automatic adjustment function in the embodiments of the present application, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be repeated.

[0177] Reference Figure 4 , which is a schematic diagram of a transcranial alternating current stimulation system with automatic adjustment function provided by an embodiment of the present application, the system includes: an acquisition module 10, a processing module 20 and an output module 30;

[0178] The acquisition module 10 is used to obtain multi-channel EEG signals distributed across multiple brain regions of the target user, and perform compressed sensing sampling on the multi-channel EEG signals based on the characteristic frequency bands of each brain region to generate compressed data;

[0179] The processing module 20 is used to perform multi-channel EEG reconstruction on the compressed data, and extract target EEG characteristic parameters corresponding to the target state in combination with physiological priors of each brain region;

[0180] The output module 30 determines the transcranial alternating current stimulation output parameters for each target brain region based on the target brain wave characteristic parameters, and outputs the signals;

[0181] The output module 30 is further configured to divide the transcranial alternating current stimulation output parameters into prior stimulation parameters and subsequent stimulation parameters;

[0182] The prior stimulation parameters are used to pre-stimulate the target brain area and determine the preliminary phase information of the brain waves of the target brain area; the subsequent stimulation parameters are dynamically adjusted based on the preliminary phase information.

[0183] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

Claims

1. A transcranial alternating current stimulation system with automatic adjustment function, characterized in that: include: Acquisition module, processing module and output module; The acquisition module is used to obtain multi-channel EEG signals distributed across multiple brain regions of the target user, and perform compressed sensing sampling on the multi-channel EEG signals based on the characteristic frequency bands of each brain region to generate compressed data; The processing module is configured to construct an adjacency matrix based on the placement of scalp electrodes or the functional division of brain regions to represent the association relationship between adjacent brain region channels; introduce a penalty term corresponding to the adjacency matrix into the reconstruction optimization objective, so that when the phase or amplitude difference between the reconstructed signals of each pair of adjacent channels exceeds a preset range, the penalty value is increased, and the penalty value is incorporated into the objective function together with the data consistency term and the sparse regularization term to obtain the reconstructed signal of each channel; Based on the preset brain region function priors, the reconstructed signal of each brain region is subjected to the first round of frequency band screening. Combined with the user's individualized priors established in baseline acquisition or historical records, the user's typical peak frequency or phase is located within the target frequency band obtained in the first round of screening, thereby extracting the target brain wave characteristic parameters corresponding to the target state; The output module determines the transcranial alternating current stimulation output parameters for each target brain region based on the target brain wave characteristic parameters, and outputs the signals; The output module is further configured to divide the transcranial alternating current stimulation output parameters into prior stimulation parameters and subsequent stimulation parameters; The prior stimulation parameters are used to pre-stimulate the target brain area and determine the preliminary phase information of the brain waves of the target brain area; the subsequent stimulation parameters are dynamically adjusted based on the preliminary phase information.

2. The transcranial alternating current stimulation system with automatic adjustment function according to claim 1, characterized in that: Also includes: When performing the multi-channel EEG reconstruction, detecting a reconstruction difference between any two channels marked as adjacent according to the adjacency matrix; When the reconstruction difference exceeds a preset reconstruction difference threshold, the measurement matrix is ​​adjusted to obtain new compressed data in a subsequent iteration, and the multi-channel EEG reconstruction is continued based on the new compressed data.

3. The transcranial alternating current stimulation system with automatic adjustment function according to claim 1 or 2, characterized in that: When performing multi-channel EEG reconstruction on the compressed data, the method includes: Calculating a graph Laplacian based on the adjacency matrix, and incorporating the graph Laplacian into a reconstruction optimization objective; During the iterative solution process, the graph Laplacian operator is used to constrain the differences between the channels so that the reconstructed signals of adjacent channels remain locally consistent in phase or amplitude.

4. The transcranial alternating current stimulation system with automatic adjustment function according to claim 3, characterized in that: Also includes: When performing graph Laplacian-constrained multi-channel EEG reconstruction, detecting local differences between channels marked as adjacent to the adjacency matrix; In response to determining that the local difference corresponds to a true brain region feature rather than noise interference, reducing the constraint weight of the graph Laplacian operator to allow the corresponding channel to maintain a reconstruction result different from that of the adjacent channel; In response to determining that the local difference belongs to noise interference, the constraint weight of the graph Laplacian operator is maintained or increased to reduce the reconstruction deviation between the interfered channel and the adjacent channel.

5. The transcranial alternating current stimulation system with automatic adjustment function according to claim 1, characterized in that: The processing module is further configured to: Based on the preset brain region function priors, corresponding target frequency bands are set for the frontal lobe, parietal lobe, and occipital lobe respectively; In response to the difference between the individualized prior and the functional prior, the frequency band interval of the corresponding brain region is locally expanded.

6. The transcranial alternating current stimulation system with automatic adjustment function according to claim 5, characterized in that: Also includes: In response to detecting a strong peak or phase that is not covered by the individualized prior and the functional prior, marking the corresponding frequency band as a suspicious interval and increasing the sampling density or iteration weight in the next reconstruction cycle; In response to repeated detection of a strong peak or a phase that persists and is independent of noise interference, the peak range corresponding to the corresponding brain region in the individualized prior is updated.

7. The transcranial alternating current stimulation system with automatic adjustment function according to claim 1, characterized in that: Also includes: When target brain wave abnormalities are detected in multiple brain regions and the association weights of any two brain regions in the adjacency matrix are greater than a preset threshold, generating a coupling abnormality brain region group; According to the target brain wave characteristics of each brain region in the abnormally coupled brain region group, respectively determining and storing the corresponding initial parameters of the prior stimulation, including stimulation frequency, initial phase and pre-stimulation duration; Simultaneously, applying a prior stimulation to the abnormally coupled brain region group and generating a phase-locking index, and if the phase-locking indexes are all greater than or equal to a first threshold after the prior stimulation ends, outputting identification information indicating that a subsequent stimulation phase has begun; In the post-stimulation stage, personalized stimulation intensity or frequency parameters for each brain region are calculated and stored, and closed-loop stimulation output is performed on each brain region based on the frequency parameters.

8. The transcranial alternating current stimulation system with automatic adjustment function according to claim 7, characterized in that: Also includes: In the post-stimulation phase, the phase difference or energy difference of each brain region in the abnormally coupled brain region group is acquired in real time, and a coupling degree measurement result is generated and updated; When the target brain wave amplitude difference or phase deviation of any brain region is detected to be lower than a second threshold, the stimulation intensity or frequency of the brain region is downgraded based on the current coupling measurement result, and the downgrade parameters are stored for subsequent iterations; When it is monitored that the phase or energy shift of the abnormally coupled brain region group occurs again as a whole, the downshift parameter and the coupling degree measurement result are used to determine whether to regenerate the initial parameters of the previous stimulation.

Images