Closed-loop adaptive transcranial electrical stimulation device and method

Abstract

The application provides a closed-loop adaptive transcranial electrical stimulation device and method, the device comprising a transcranial electrical stimulation unit and an electroencephalogram acquisition unit controlled by a control unit and sharing the same set of electrodes, and the control unit controls the connection of the electrodes through a switching circuit. The method acquires electroencephalogram data and paradigm behavior data, combines multi-modal data fusion calculation to evaluate the state of the subject, compares it with the preset control target, adopts the preset control model for control, and completes the state evaluation result of the subject and the preset control target without deviation. The application can realize a closed-loop adaptive transcranial electrical stimulation control mode.

Description

Technical Field

[0001] The present invention relates to a brain state regulation method, and in particular to a closed-loop adaptive transcranial electrical stimulation device and method that integrates high-precision transcranial electrical stimulation (HD-tDCS), electroencephalogram (EEG) and behavioral paradigms. Background Art

[0002] Transcranial electrical stimulation (tES) uses external electrodes to apply low-intensity currents in specific patterns directly to specific brain regions to modulate neural activity. It is an important form of noninvasive neuromodulation. TES techniques include transcranial direct current stimulation (DCS), transcranial alternating current stimulation (ACS), and transcranial pulse stimulation (PPS). TES neuromodulation technology is applicable to both non-medical and medical settings. Non-medical settings primarily include enhancing brain function for specialized populations such as military personnel, armed police, firefighters, and astronauts, as well as brain science research conducted by scientific researchers. Medical settings include neuromodulation for patients with neurological and psychiatric conditions such as stroke rehabilitation, Alzheimer's disease, depression, and aphasia. While the principles of these two settings are generally the same, the specific evaluation training paradigms, brain state assessment methods, and neuromodulation parameters may differ. Traditional TES devices can only perform TES, and the effects of stimulation can only be observed and evaluated through empirical observation. However, with the increasing demands of scientific research and clinical practice, the digital and precise evaluation of brain state changes using TES is gaining attention.

[0003] Before implementing transcranial electrical stimulation, brain state assessment technology is required. Brain state assessment technology can generally be achieved through physiological signal detection and behavioral paradigm detection. Among them, brain state assessment technology based on electroencephalogram (EEG) is a commonly used method for physiological signal detection and assessment of brain state. Behavioral paradigm detection generally characterizes brain state through the accuracy of behavioral performance and reaction time. Accurate digital assessment of brain state assessment is generally based on norm-established EEG algorithms and behavioral result algorithm assessment for accurate data characterization, but accurate assessment based on the integration of EEG and behavioral results is still a gap.

[0004] Traditional transcranial electrical stimulation requires the use of other EEG equipment and behavioral paradigm software to evaluate the stimulation effect, and then the transcranial electrical stimulation mode is adjusted manually based on the evaluation results. This method is cumbersome to operate, the operation process is lengthy, and it is prone to errors, which cannot adapt to the development requirements of intelligent neural regulation. Furthermore, traditional transcranial electrical stimulation technology has two relatively important limitations. First, traditional sponge electrodes can only achieve large-area stimulation of the scalp and cannot accurately control the stimulation area, which has limitations for the treatment of many diseases. Second, stimulation is mostly provided in an "open-loop" manner, that is, it is adjusted according to the pre-programmed active setting of stimulation parameters, and it is impossible to adjust and feedback the clinical symptoms or disease evaluation effects that appear in patients after stimulation.

[0005] Traditional brain state assessment techniques typically separate physiological signal evaluation from paradigm assessment, lacking an integrated assessment algorithm. This traditional approach fails to produce comprehensive evaluation results and provides precise data guidance for closed-loop control. Current closed-loop control models are primarily offline, meaning that brain state assessment is performed offline after transcranial electrical stimulation (TES) is completed, and the TES pattern is further adjusted based on the results. Even with online closed-loop control, traditional methods often separate the acquisition and stimulation devices, with the acquisition and stimulation modules synchronized via an external system. This closed-loop control model fails to form a cohesive system, resulting in poor synchronization, complex setup, and difficult operation.

[0006] It can be seen that if the existing technology wants to achieve the above functions, it requires "transcranial electrical stimulation equipment, EEG acquisition equipment, task paradigm display equipment and personnel response (reaction time, response accuracy) recording equipment, and synchronizers for the above three devices." A total of four independent sets of hardware and one set of post-statistical software are required, and then a professional will make a judgment based on the statistical results. In addition, because the previous system is "one-way timestamping": that is, the transcranial electrical stimulation time, paradigm task time, and personnel response time are all synchronized to the computer system to form a data packet for post-analysis; however, it cannot fully automatically closed-loop "reverse control of stimulation mode and reverse regulation of task difficulty" based on EEG changes and personnel behavioral response changes; "stimulation mode and task difficulty" can only be manually adjusted by professionals. Summary of the Invention

[0007] The application provides a closed-loop adaptive transcranial electrical stimulation device and method, relates to a transcranial electrical stimulation nerve regulation method and a brain state evaluation method, aims at the disadvantage that a traditional transcranial electrical stimulation device cannot accurately evaluate and automatically feedback control stimulation effect, solves the brain state regulation problem of integration of transcranial electrical stimulation, electroencephalogram acquisition and paradigm operation, solves the construction problem of the brain state regulation device, and realizes a high-precision transcranial electrical stimulation (HD-tDCS), EEG acquisition and behavior paradigm integrated hardware system and software system.

[0008] A brain electrical signal acquisition and transcranial electrical stimulation integrated nerve regulation device comprises a transcranial electrical stimulation unit and a brain electrical signal acquisition unit controlled by a control unit, the transcranial electrical stimulation unit and the brain electrical signal acquisition unit share the same set of electrodes through a switching circuit, and the switching circuit is controlled by the control unit; the transcranial electrical stimulation unit comprises a positive and negative high-voltage circuit, the positive and negative high-voltage circuit is connected to the electrodes through the switching circuit, the brain electrical signal acquisition unit comprises a filter circuit and an ADC acquisition circuit connected to each other, and the filter circuit is connected to the electrodes through the switching circuit.

[0009] The electrode comprises an electrode cup cover, a sheet-shaped electrode and an electrode cup, the electrode cup cover is fastened to the upper surface of the electrode cup, the bottom of the electrode cup is provided with an annular support structure for placing the sheet-shaped electrode, and the lower part of the electrode cup cover is provided with an electrode cup foot which presses the upper end of the sheet-shaped electrode.

[0010] The brain electrical signal acquisition unit comprises a brain electrical signal acquisition circuit connected to the electrodes, and the brain electrical signal acquisition circuit uses an AD7768 chip or an ADS1299 chip for 8-channel synchronous sampling.

[0011] The brain electrical signal acquisition circuit adopts daisy-chain cascading, and 8 channels are formed into 16 channels, 32 channels, 64 channels and 128 channels.

[0012] The brain electrical signal acquisition circuit leads out all positive and negative electrodes, and differentially acquires brain electrical signals between the positive and negative electrodes; or the negative end is internally connected together, and brain electrical signals are differentially acquired between the positive electrode and the unified reference.

[0013] The transcranial electrical stimulation unit comprises a high-voltage generation circuit and a counter-pressure circuit connected to the high-voltage generation circuit, and the high-voltage generation circuit adopts a tps61230 voltage boosting chip.

[0014] The transcranial electrical stimulation circuit is further provided with a constant current feedback detection circuit, the constant current feedback detection circuit sends a stimulation current value of the positive and negative high-voltage circuit to the control unit, the constant current feedback detection circuit is provided with a second amplifier, and the second amplifier adopts an SGM8622.

[0015] The electroencephalogram acquisition unit further comprises a lead-off judgment circuit, which is located between the filtering circuit and the switching circuit and is used for judging whether the electrode is off by using a hysteresis comparator circuit.

[0016] The control unit is provided with a processor, which is an STM32F383CC chip.

[0017] A closed-loop adaptive transcranial electrical stimulation method comprises the following steps:

[0018] S1: different test paradigms are selected according to different subjects;

[0019] S2: a preset control target is set as a level setting for brain state evaluation after the subject is controlled;

[0020] S3: a preset control model is set to determine the output parameters of transcranial electrical stimulation;

[0021] S4: electroencephalogram data are collected and processed to obtain the evaluation result of the subject's electroencephalogram state;

[0022] S5: behavioral data are collected and evaluated according to the paradigm;

[0023] S6: according to the electroencephalogram precise evaluation algorithm and the paradigm evaluation algorithm, a fusion brain state evaluation algorithm is generated to calculate the evaluation result of the subject's state;

[0024] S7: the deviation between the evaluation result of the subject's state and the preset control target is calculated;

[0025] S8: the control model is called to control the output parameters of transcranial electrical stimulation according to the deviation value;

[0026] S9: steps S3-S8 are repeated to cycle online evaluation and test control until there is no deviation between the evaluation result of the subject's state and the preset control target.

[0027] Further, in step S2, the control target is divided into ten levels, including five stages of severe positive deviation, positive deviation, basically no deviation, negative deviation and severe negative deviation, and each stage is divided into high and low grades.

[0028] Further, in step S3, the setting steps of the output parameters are as follows:

[0029] S11: the mode of transcranial electrical stimulation is determined;

[0030] It is divided into transcranial direct current stimulation, transcranial alternating current stimulation, transcranial pulse stimulation and transcranial random stimulation mode;

[0031] S2: the position of transcranial electrical stimulation is determined;

[0032] The position is selected in the international standard electroencephalogram position distribution map, and is provided with 1 to N electric stimulation positions, wherein the positive pole is 1 to N-1 positions, and the negative pole is 1 to N-1 positions, which jointly constitute a regulated circuit combination, wherein N is a natural number, and is greater than or equal to 2;

[0033] S3: determining specific parameters of transcranial electric stimulation;

[0034] including the polarity of the stimulation current, the stimulation current duration, the stimulation current amplitude, the stimulation current bias, the stimulation current frequency, and the stimulation current duty cycle;

[0035] S4: determining the evaluation mode of the control target;

[0036] The evaluation mode of the control target is divided into asynchronous and synchronous modes, wherein the asynchronous mode is to evaluate whether the control target is achieved after the control is executed, and the synchronous mode is to evaluate whether the control target is achieved while the control is being performed;

[0037] S5: determining the operation mode when the control target is achieved or not achieved;

[0038] When the control target is achieved, the current control is stopped, or the control is continued according to a fixed time, so as to consolidate the target; when the control target is not achieved, the control can be continued according to a set time, and the control is stopped again regardless of whether the control target is achieved or not; or the control is directly stopped when the control target is not achieved.

[0039] Further, in step S4, the electroencephalogram state evaluation includes data preprocessing, related feature analysis and extraction, machine learning, and obtaining the electroencephalogram state evaluation result of the subject;

[0040] The data preprocessing is to remove bad data or data full of artifacts without changing clean data, or to apply filters or spatial transformation to change clean data to facilitate analysis, including re-reference, filtering, and independent principal component analysis processing;

[0041] The related feature analysis and extraction are used to extract the basic data of electroencephalogram analysis, including Fourier transform, nonlinear transform, time domain analysis, and brain network analysis processing;

[0042] The machine learning and obtaining the electroencephalogram state evaluation result of the subject realize the classification task in supervised learning, including support vector machine, neural network framework, and auto-encoding network processing mode.

[0043] Further, in step S5, the paradigm behavior state evaluation includes: statistics of the reaction time of the subject, statistics of the reaction accuracy of the subject, and calculation of the reaction time and reaction accuracy characteristics of the subject.

[0044] Furthermore, in step S6, the accuracy reaction time of the deviation state and the non-deviation state is introduced into the discriminant model as a behavioral indicator:

[0045] The first method is to perform feature fusion based on data of different modalities. The fusion methods include data-level fusion, decision-level fusion, and intermediate fusion.

[0046] The second method is to not consider the accuracy response time as data of different modalities, but only use it as a general feature and normalize it with the EEG features;

[0047] The third method is to perform algorithm fusion of different data modalities based on the differences between the absolute control group populations.

[0048] In the first method:

[0049] The data level fusion aims to fuse multiple independent data into a single feature vector, which is then input into the classifier. Specific methods include concatenation, bitwise multiplication, and bitwise addition.

[0050] The decision-level fusion aims to train classifiers separately using data from different modalities, and then score the outputs to fuse the results. Specific methods include maximum fusion, average fusion, Bayesian rule fusion, and ensemble learning.

[0051] The intermediate layer fusion is to convert different modal data into high-dimensional feature expressions and then fuse them with the intermediate layer of the model.

[0052] Furthermore, in step S8, if the evaluation result meets the control target, the control may not be started; or the same level consolidation control mode may be started in a preset manner.

[0053] Furthermore, in step S5, paradigm evaluation includes but is not limited to the Posner attention paradigm and the n-back working memory paradigm.

[0054] The closed-loop adaptive transcranial electrical stimulation device and method provided by the present invention are used to automatically evaluate the brain state and automatically match different precise control modes, thereby realizing a closed-loop adaptive transcranial electrical stimulation control mode.

[0055] The closed-loop adaptive transcranial electrical stimulation device and method include high-precision transcranial electrical stimulation, integrated brain electrical acquisition and transcranial electrical stimulation technology, acquisition and stimulation compatible electrode, neural state analysis and evaluation model based on brain electrical and paradigm fusion, and precise neural regulation model. Among them, the high-precision transcranial electrical stimulation can effectively lock the stimulation area and realize precise regulation. The integrated brain electrical acquisition technology solves the key technical problems of multi-level protection, multi-level switching and multi-level isolation of the circuit. The acquisition and stimulation compatible electrode can adapt to the current and impedance requirements of the acquisition and stimulation two different scenes, and realize the two different operation functions of signal acquisition and stimulation of the same electrode. The neural state analysis and evaluation model establishes a reliable algorithm model through pattern recognition and machine learning based on specific scene requirements, providing data guidance for closed-loop regulation. The precise neural regulation model sets differential neural regulation parameters based on different evaluation results.

[0056] The closed-loop adaptive transcranial electrical stimulation device and method is an integrated mode of neural regulation, which forms a closed-loop adaptive linkage of acquisition, stimulation and evaluation functions, fully develops the potential and advantages of transcranial electrical stimulation technology, and can dynamically adjust the intelligent adaptive treatment mode of neural regulation according to the brain state evaluation results. It provides intuitive diagnosis and treatment data guidance for special personnel, scientific researchers and clinicians, greatly improves the feedback efficiency of evaluation analysis and regulation effect, and enhances the pertinence and effectiveness of regulation. It is a new type of intelligent neural regulation method. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 It is a frame schematic diagram of the brain electrical acquisition and transcranial electrical stimulation integrated neural regulation device;

[0058] Figure 2 It is a structural schematic diagram of the brain electrical electrode;

[0059] Figure 3 It is a schematic diagram of the ADC acquisition circuit of the brain electrical acquisition unit;

[0060] Figure 4 It is a schematic diagram of the high-voltage generation circuit;

[0061] Figure 5 It is a schematic diagram of the back pressure circuit;

[0062] Figure 6 It is a schematic diagram of the constant current feedback detection circuit;

[0063] Figure 7 It is a flowchart of the closed-loop adaptive transcranial electrical stimulation method;

[0064] Figure 8 It is a flowchart of the brain electrical acquisition;

[0065] Figure 9 is a flow chart of the paradigm operation system;

[0066] Figure 10 is a schematic diagram of the Posner attention paradigm;

[0067] Figure 11 is a schematic diagram of the 2-back paradigm. DETAILED DESCRIPTION

[0068] like Figure 1 As shown, the closed-loop adaptive transcranial electrical stimulation device includes a transcranial electrical stimulation unit and an EEG acquisition unit controlled by a control unit. The transcranial electrical stimulation unit and the EEG acquisition unit share the same set of electrodes through a switching circuit, and the switching circuit is controlled by the control unit; the transcranial electrical stimulation unit includes positive and negative high-voltage circuits, which are connected to the electrodes through a switching circuit, and the EEG acquisition unit includes a filter circuit and an ADC acquisition circuit connected to each other, and the filter circuit is connected to the electrodes through a switching circuit.

[0069] The EEG acquisition unit also includes a lead-off judgment circuit, which is located between the filter circuit and the switching circuit and is used to judge whether the electrode has fallen off. The lead-off judgment circuit is an existing technology and can be implemented using a hysteresis comparator circuit.

[0070] The control unit is provided with a processor, which adopts an STM32F383CC chip and is connected to a computer to realize data transmission.

[0071] like Figure 2 As shown, the electrode includes an electrode cup cover 1, a sheet electrode 2 and an electrode cup 3. The electrode cup 3 is mounted on an electrode cap 7. The electrode cup cover 1 is buckled on the upper surface of the electrode cup 3. The bottom of the electrode cup 3 is provided with an annular support structure 8 for placing the sheet electrode 2. The electrode cup cover 1 is provided with an electrode cup foot 10 for cooperating with the annular support structure 8 to fix the sheet electrode 2. The bottom of the sheet electrode 2 is used to apply EEG paste 5. The top of the sheet electrode 2 is connected to an electrode wire 6. By vertically extending the wire, the sheet electrode 2 can be lowered to the bottom of the electrode cup 3, reducing the distance between the sheet electrode 2 and the head 9, thereby improving the quality of EEG signal acquisition and the effect of transcranial electrical stimulation. In this embodiment, the multiple electrodes are mounted on the electrode cap through an electrode holder, and the electrode cap is worn on the head of the user.

[0072] like Figure 3 As shown, the ADC acquisition circuit is connected to the control unit. The ADC acquisition circuit of the EEG acquisition unit can use AD7768, ADS1299 chips, or use op amp + high-precision ADC to collect electrode signal data.

[0073] The AD7768 and ADS1299 chips used in the embodiment are 8-channel synchronous sampling chips, which can be cascaded through a daisy chain to form 16-channel, 32-channel, 64-channel, and 128-channel electroencephalogram acquisition circuits. The electroencephalogram acquisition unit can lead out the positive and negative poles of the electrodes to differentially acquire electroencephalogram signals between the positive and negative poles; or the negative poles of the electrodes can be internally connected together, and the electroencephalogram signals are differentially acquired between the positive pole and the unified reference (RFE ELEC).

[0074] As shown in Figure 4 and Figure 5 , the positive and negative high-voltage circuit includes a high-voltage generation circuit and a reverse voltage circuit connected to the high-voltage generation circuit. The high-voltage generation circuit uses a tps61230 voltage boosting chip. The reverse voltage circuit is a prior art and will not be described again. The positive and negative high-voltage circuit realizes free switching of cathode stimulation and anode stimulation through positive and negative voltages.

[0075] The conventional voltage boosting circuit generates a fixed high voltage through resistance division. The high-voltage generation circuit dynamically controls the specific value of the high voltage through a single IO pin to realize an output of 5-20V. The reverse voltage circuit is a prior device, which realizes an effect of inputting a few volts and outputting a negative few volts.

[0076] The impedance between the positive and negative electrodes will have a large difference due to different positions of the electrodes on the human head, different treatments of the human scalp grease, and different treatments of the electroencephalogram paste. The high voltage required by the transcranial electrical stimulation unit is different due to different impedances. For example, if the impedance between the positive and negative electrodes is 3KΩ, and the current is 2mA, only a 6V voltage difference between the positive and negative electrodes is needed; if the impedance between the positive and negative electrodes is 10KΩ, and the current is 2mA, a 20V voltage difference between the positive and negative electrodes is needed. The battery is generally a 4.2V lithium battery. Only by dynamically adjusting the output high voltage range according to the impedance requirement, can the transcranial electrical stimulation unit have a high efficiency.

[0077] As shown in Figure 6 , the transcranial electrical stimulation circuit further has a constant current feedback detection circuit, which sends the stimulation current value of the positive and negative high-voltage circuit to the control unit. The constant current feedback detection circuit has a second amplifier, which uses an SGM8622. The positive input end and the negative input end of the second amplifier are connected with a high-precision resistor R66. In use, the stimulation current from the positive and negative high-voltage circuit is differentially sampled through the high-precision resistor R66 to become a single-ended signal, which is then input to a high-precision ADC. After sampling by the ADC, the processor controls the positive and negative high-voltage circuit to make dynamic adjustment within a certain range by comparing the deviation between the set current and the actual current.

[0078] The switching circuit is controlled by the control unit, and the working mode of the electrode is switched, because the electrode can realize transcranial electrical stimulation and electroencephalogram acquisition, but cannot realize transcranial electrical stimulation and electroencephalogram acquisition at the same time, so the electrode can be connected to the transcranial electrical stimulation unit or the electroencephalogram acquisition unit through the switching circuit at the same time, and the switching circuit can be realized by using an electronic switching switch.

[0079] The working mode of the electrode for transcranial electrical stimulation or electroencephalogram acquisition is processed by the control unit, when the control unit controls the positive and negative high-voltage circuit to send a stimulating current, the switching circuit switches the working mode of the electrode to the working mode of electrical stimulation; when the control unit controls the positive and negative high-voltage circuit to stop sending a stimulating current, the switching circuit switches the working mode of the electrode to the working mode of electroencephalogram acquisition.

[0080] In use, the working mode of the plurality of electrodes can use a synchronous control mode and an asynchronous control mode, the synchronous control mode refers to transcranial electrical synchronous stimulation-electroencephalogram synchronous acquisition, the electrode simultaneously performs transcranial electrical stimulation, or the electrode simultaneously performs electroencephalogram acquisition, and can be made into 32 channels-128 channels. The asynchronous control mode refers to transcranial electrical synchronous stimulation-electroencephalogram asynchronous, the electrode performs electroencephalogram acquisition or the electrode performs transcranial electrical stimulation, and cannot be performed at the same time, and can be made into 32 channels-128 channels. Figure 8 Asynchronous control mode is represented in the figure.

[0081] As Figure 7 As shown in the figure, the closed-loop adaptive transcranial electrical stimulation method is realized by a closed-loop adaptive transcranial electrical stimulation device, and includes the following steps:

[0082] S1: Different test paradigms are selected according to different subjects;

[0083] First, the subjects are determined according to soldiers, armed police, firefighters, ordinary young people, the elderly, children and the like; different test paradigms are selected according to different subjects, so that the pertinence and adaptability are improved.

[0084] S2: A preset control target is set as a level setting of brain state evaluation after control of the subject;

[0085] The brain state evaluation after control is divided into ten levels, and a total of five stages of serious positive deviation, positive deviation, basically no deviation, negative deviation and serious negative deviation, each stage is divided into high and low grades, thereby forming a level setting of ten levels.

[0086] A control level is determined in the ten levels as a control target, in principle, the control target should be set step by step according to the result of the brain state evaluation of the subject last time. In special cases, the control target can also be set across levels. Similarly, the control target can also be consolidated without crossing levels.

[0087] S3: Preset the control model and determine the output parameters of transcranial electrical stimulation;

[0088] According to different brain state control targets, determine whether to control positively or negatively, and according to different brain state control targets, determine the specific control parameters. The order of setting the output parameters is as follows:

[0089] (1) Determine the mode of transcranial electrical stimulation;

[0090] It is divided into several modes, including transcranial direct current stimulation, transcranial alternating current stimulation, transcranial pulse stimulation, and transcranial random stimulation.

[0091] (2) Determine the location of transcranial electrical stimulation;

[0092] The positions can be selected from the international standard EEG position distribution map, and can be electrically stimulated from 1 to N. Among them, the positive pole can be at most N-1 positions, and the negative pole can be at most N-1 positions, which together constitute the circuit combination for regulation.

[0093] (3) Determine the specific parameters of transcranial electrical stimulation;

[0094] These include parameters such as stimulation current polarity, stimulation current duration, stimulation current amplitude, stimulation current bias, stimulation current frequency, and stimulation current duty cycle. Stimulation current polarity includes both positive and negative polarity; stimulation duration ranges from 1 minute to 6-0 minutes; stimulation current amplitude ranges from 300vA to 2000vA; stimulation bias current + stimulation current amplitude can be set within ±2000vA; stimulation frequency ranges from DC to 250Hz; and stimulation current duty cycle ranges from 0% to 100%, with amplitude from DC to 250Hz.

[0095] (4) Determine the evaluation method for regulatory objectives;

[0096] There are two ways to evaluate control targets: asynchronous and synchronous. The asynchronous method evaluates whether the control target has been achieved after the control is completed, while the synchronous method evaluates whether the control target has been achieved while the control is being executed.

[0097] (5) Determine the operating mode when the control objectives are achieved or not achieved;

[0098] When the control target is achieved, you can stop the current control, or continue to control according to a fixed time to consolidate the target.

[0099] If the control target is not reached, the control can be continued for a set time, and then stopped regardless of whether the control target is reached. Or the control can be stopped directly if the control target is not reached.

[0100] S4: Collecting EEG data and processing, obtaining the subject EEG state evaluation results;

[0101] The EEG state evaluation process includes: data preprocessing, relevant feature analysis and extraction, machine learning, and obtaining the subject EEG state evaluation results.

[0102] S5: Collecting behavioral data and evaluating the paradigm;

[0103] The paradigm evaluation algorithm includes: (1) statistics of the subject reaction time, (2) statistics of the subject reaction accuracy, and according to (1) and (2) to calculate the reaction time and reaction accuracy of the subject within and between subjects.

[0104] S6: According to the EEG data, behavioral data, form fusion calculation algorithm, so as to evaluate the "subject state evaluation results";

[0105] Since there are both behavioral indicators and EEG indicators to judge the state of the test personnel, they need to be combined for comprehensive evaluation. Specifically, the accuracy rate reaction time of deviation state and non-deviation state can be introduced into the deviation discrimination model as a behavioral indicator.

[0106] The first method is to fuse the features of different modalities. The basic fusion methods include data level fusion, decision level fusion, and intermediate fusion.

[0107] Data level fusion: aims to combine EEG state evaluation result data and behavioral paradigm result data into a single feature vector, and then input it into the classifier. The specific methods include splicing, bitwise multiplication, and bitwise addition. Since the accuracy rate reaction time is a one-dimensional feature, bitwise multiplication and bitwise addition are not suitable, so splicing can be tried. However, compared with high-dimensional EEG data, this one-dimensional feature may be submerged in high-dimensional data features, and its complementary effect may not be obvious.

[0108] Decision level fusion: aims to train classifiers using EEG state evaluation result data and behavioral paradigm result data respectively, and then fuse the outputs (decisions) to obtain the result. The specific methods include maximum fusion, average fusion, Bayesian rule fusion, and ensemble learning. The specific method to be used needs to be determined through numerical experiments.

[0109] Intermediate layer fusion: mainly converts EEG state evaluation result data and behavioral paradigm result data into high-dimensional feature representation, and then fuses them with the intermediate layer of the model. This method is mainly used for feature fusion in neural networks. For pilot data, EEG high-dimensional features can be mapped to spatially symmetric manifolds, and accuracy rate reaction time can be mapped to one-dimensional manifolds. Data fusion on the two manifolds can then be performed before entering the classifier.

[0110] The second method is not to regard the accuracy reaction time of the behavioral paradigm as data of different modalities, but only to treat it as a general feature and normalize it with the EEG discrimination feature. This operation is simple, but the role of the accuracy reaction time may be weakened.

[0111] The third method uses the differences between the target and control groups to perform algorithmic fusion of different data modalities. Specifically, the target and control groups are evaluated using identical experimental methods. The difference between the EEG neural state evaluation results and behavioral evaluation results such as accuracy and reaction time is then calculated on a percentage basis. Greater differences are weighted, while smaller differences are weighted, thus forming an objective fusion weight ratio.

[0112] S7: Calculate the deviation between the subject's state evaluation result and the control target;

[0113] The comprehensive evaluation algorithm in the above steps is used to comprehensively evaluate the subject's status level, and then compared with the level requirements of the preset control target to determine the level gap between the subject's status evaluation result and the preset control target.

[0114] S8: calling the control model to adjust the transcranial electrical stimulation output parameters according to the deviation value;

[0115] Based on the "subject status evaluation results" in step S6, the gap between the evaluation level and the control target is assessed, and then the preset control model is called to perform the control. In particular, if the evaluation results already meet the control target, the control can be disabled, or the same level consolidation control mode can be activated according to the preset method.

[0116] The control model uses the output parameters of transcranial electrical stimulation in step S3 above. The output parameters of transcranial electrical stimulation include: stimulation location, stimulation time, output stimulation mode (DC stimulation, AC stimulation, pulse stimulation, random stimulation, depth stimulation), output stimulation frequency, output stimulation polarity, output stimulation intensity, and output start / stop.

[0117] S9: Repeat steps S3-S6, and circulate the online evaluation test control until there is no deviation between the "subject status evaluation result and the control target".

[0118] Among them, the evaluation methods can be divided into synchronous evaluation and asynchronous evaluation.

[0119] Furthermore, in step S4, Figure 8 As shown, the EEG state evaluation includes data preprocessing, relevant feature analysis and extraction, machine learning, and obtaining the EEG state evaluation results of the subjects. The details are as follows:

[0120] S41: Data Preprocessing: Preprocessing aims to remove bad data or artifact-ridden data without changing the clean data, or to apply filters or spatial transformations to change the clean data to make it easier to analyze. The main methods for processing EEG data include:

[0121] (1) Re-reference: The potential of a certain electrode measured by EEG. However, since the potential of a single electrode cannot be measured, an electrode with a known potential (preferably zero potential) is required as a reference electrode. The potential of an ideal reference electrode should be zero (its potential should not change). Currently, the more common reference methods include bilateral mastoid reference, average reference, and bipolar reference.

[0122] (2) Filtering: Eliminate high-frequency artifacts and low-frequency drift; filter out different frequency bands depending on the analysis task and data characteristics. The frequency band used in the analysis is usually between 0.5Hz and 70Hz, and band-stop filtering is required to eliminate mains interference.

[0123] (3) Independent principal component analysis: This can be used to clean EEG data by identifying components that separate artifacts and then subtracting them from the data. Dimensionality reduction can also be achieved by analyzing the component time series. Components related to blinking and eye muscle closure can be removed by independent component analysis.

[0124] S42: Related feature analysis and extraction include: Fourier transform, nonlinear transformation, time domain analysis, and brain network analysis.

[0125] (4) Fourier transform: Observe the changes in brain state, the corresponding energy change characteristics of waves of different frequencies, and extract the ones with large contrast ratios as distinguishing features.

[0126] (5) Nonlinear indicators: Entropy (En) is a physical quantity that can characterize the complexity of EEG. Entropy-related features are extracted, such as Fisher information (FI), spectral entropy (SpEn), Shannon's entropy (ShEn), and approximate entropy (ApEn). In addition, sample entropy and permutation entropy can also be used as features.

[0127] (6) Time domain analysis: The EEG recordings are observed and analyzed using the properties of the EEG waveform, such as amplitude, mean, variance, skewness, and kurtosis. In EEG signal research, commonly used time domain analysis methods include: zero-crossing analysis, histogram analysis, variance analysis, correlation analysis, peak detection, waveform parameter analysis, and waveform recognition.

[0128] (7) Brain network analysis: Power-based connectivity: Analyzes the time-frequency power between two electrodes across time, calculating the correlation between activities at the same or different frequencies and at the same or different time points. Mutual information: Calculated based on the distribution of values ​​within a variable and the joint distribution of two (or more) variables, detecting shared information between two variables.

[0129] S43: binary or multi-classification machine learning to obtain a multi-dimensional and multi-quality discrimination model for EEG analysis of the subjects;

[0130] This step is a supervised learning task. In supervised learning, an algorithm learns from labeled data. After understanding the data, the algorithm determines the appropriate label for the new data by associating patterns with new, unlabeled data. Supervised learning primarily involves classification and regression. Classification is a technique for determining the category of a dependent variable based on one or more independent variables. Regression problems predict numerical values ​​based on previously observed data. Therefore, this task falls under the classification category within supervised learning.

[0131] Common classification tasks include binary classification and multi-classification. For example, determining fatigue status is a binary classification task, while determining fatigue level is a multi-classification task. Classification tasks are typically implemented using traditional classifier machine learning algorithms and neural network deep learning algorithms.

[0132] (1) Support Vector Machine (SVM): A support vector machine is a generalized linear classifier that performs binary classification on data using supervised learning. Its decision boundary is the maximum margin hyperplane obtained by solving the learning samples. When performing linear classification, the classification surface is located at a large distance from the two types of samples. When performing nonlinear classification, the nonlinear classification is transformed into a linear classification problem in a high-dimensional space through high-dimensional space transformation. There is a problem of choosing the kernel function in support vector machines. The commonly used kernels are linear kernels for solving linearly separable problems, and other nonlinear kernels for solving linearly inseparable problems (such as Gauss kernel function and polynomial kernel function).

[0133] (2) Neural network framework:

[0134] Fully connected neural network: A fully connected neural network model is a multilayer perceptron (MLP). The principle of a perceptron is to find the most reasonable and robust hyperplane between categories. The most representative perceptron is the support vector machine algorithm. Neural networks draw inspiration from both perceptrons and biomimetic theory. Generally speaking, after receiving a signal, animal nerves send signals to various neurons. After receiving the input, each neuron activates and generates output signals based on its own judgment. These signals are then aggregated to identify and classify the information source.

[0135] Convolutional Neural Networks (CNNs) are a type of feedforward neural network with a deep structure that incorporates convolutional computations. They are a representative algorithm for deep learning. CNNs possess representation learning capabilities and can perform shift-invariant classification of input information based on their hierarchical structure. They are therefore also known as shift-invariant artificial neural networks (SIANNs). Convolutional neural networks differ from conventional neural networks in that they incorporate a feature extractor consisting of convolutional layers and subsampling layers (pooling layers), significantly simplifying the model complexity and reducing the number of parameters.

[0136] (3) Autoencoder network: Autoencoder network is a type of unsupervised learning network that can automatically learn features from unlabeled data. It is a neural network that aims to reconstruct input information. It can give better feature descriptions than the original data and has strong feature learning capabilities. In deep learning, features generated by autoencoder networks are often used to replace original data to achieve better results.

[0137] Algorithm model construction requires consideration of multiple features and subsequent fusion. For example, in brain networks, features such as the coherence matrix, mutual information matrix, and phase lag matrix can be extracted; statistical features and nonlinear features (such as entropy) can also be extracted from the original signal.

[0138] Because the corresponding EEG signals are multi-channel signals, the features obtained are often in matrix form. Therefore, the matrix can be used as raw data to input into the neural network for training. At the same time, considering that it comes from multiple aspects such as brain networks, time-frequency domains, statistical characteristics, and nonlinear characteristics, machine learning methods can be used to first train the network and then perform fusion and classification to ultimately form the optimal algorithm model.

[0139] After the EEG data is processed as described above, the evaluation results of the subject's brain state can be obtained.

[0140] like Figure 9 The paradigm operating system shown includes various paradigms such as perception, attention, memory, expression, language, decision-making, and emotion. These paradigms all have data evaluation indicators such as accuracy and reaction time. These various paradigms can form various psychological experimental paradigms, disease diagnosis and treatment paradigms, and game paradigms.

[0141] The paradigm operation system can select the difficulty level and accurately record stimulus presentation duration, subject reaction time, and response accuracy, thereby providing basic data for the paradigm behavioral state evaluation system software. The paradigm algorithm can be determined by the percentile of a subject within the population norm. First, it is necessary to calculate the characteristics of reaction time and response accuracy within the subject; calculate the characteristics of reaction time and response accuracy between subjects; and calculate the percentile of the subject's reaction time and response accuracy within the same test population.

[0142] In step S5, the present invention may include the Posner attention paradigm (spatial paradigm) and the n-back working memory paradigm and other types of paradigms.

[0143] like Figure 10 As shown, the Posner paradigm consists of several trials. In each trial, a spatial cue stimulus (a leftward or rightward-pointing arrow) first appears in the center of the visual field, directing participants' attention implicitly (without eye movement, maintaining fixation on the center of the visual field) to the spatial location indicated by the cue. After the cue disappears, a short, randomly-lengthed delay period occurs. During this delay, participants are required to maintain spatial attention to the indicated location. After the delay, either a circular Gabor grid standard stimulus (50% probability) or a target stimulus (50% probability) appears at the indicated location (80% probability, valid stimulus) or at the location opposite to the indicated location (20% probability). Regardless of whether the stimulus appears at the valid or invalid location, participants are instructed to quickly respond by pressing a key to indicate whether the stimulus is a target. The difference between the target and standard stimuli is the density of the Gabor grid. Setting a smaller density difference makes it more difficult to distinguish between the target and standard stimuli, thereby requiring participants to engage in more attention. Accuracy and reaction time are recorded to assess attentional level and performance.

[0144] The experiment lasted a total of 5 minutes. The grating stripe angle was 60° / 120°. The waiting time before the indicator arrow was displayed was 2 seconds. The indicator arrow was displayed for 0.2 seconds. The misleading rate of the indicator arrow was 20%. The waiting time after the indicator arrow disappeared was 1 to 4 seconds. The grating display time was about 0.02 seconds. The maximum effective reaction time was 1 to 3 seconds.

[0145] like Figure 11 As shown, the n-back paradigm is a classic working memory paradigm. It requires participants to compare a recently presented stimulus with the nth stimulus preceding it. Cognitive load is manipulated by controlling the number of stimuli between the current and target stimuli. The task used was a letter matching task. This study employed a 2-back paradigm. The experiment lasted 5 minutes, with a character display duration of 0.5 seconds, a maximum effective reaction time of 2 seconds (including the letter display duration), and a 1-second wait time for the next character to appear.

[0146] Statistical analysis of the characteristics of reaction time and reaction accuracy within the subject; Statistical analysis of the characteristics of reaction time and reaction accuracy between subjects; Statistical analysis of the percentiles of reaction time and reaction accuracy of the subjects in the same test population.

[0147] The closed-loop adaptive transcranial electrical stimulation device and method provided by the present invention include high-precision transcranial electrical stimulation, integrated EEG acquisition technology, acquisition and stimulation compatible electrodes, a neural state analysis and evaluation model based on EEG and paradigm fusion, a precise neural control model, etc. Among them, high-precision transcranial electrical stimulation can effectively lock the stimulation area and achieve precise control. The integrated EEG acquisition technology solves key technical problems such as multi-level protection, multi-level switching, and multi-level isolation of circuits. The acquisition and stimulation compatible electrodes can adapt to the current and impedance requirements of two different scenarios of acquisition and stimulation, so that the same electrode can both acquire signals and perform two different operational functions of stimulation. The neural state analysis and evaluation model is based on specific scenario requirements and establishes a reliable algorithm model through pattern recognition and machine learning, providing data guidance for closed-loop control. The precise neural control model sets targeted differentiated neural control parameters based on different evaluation results.

[0148] The closed-loop adaptive transcranial electrical stimulation device and method is an integrated neural control method that forms a closed-loop adaptive linkage between the three functions of acquisition, stimulation, and evaluation, fully unleashing the potential and advantages of transcranial electrical stimulation technology. It is an intelligent adaptive treatment model that can dynamically adjust the neural control method according to the brain state evaluation results, providing intuitive diagnosis and treatment data guidance for special personnel, scientific researchers, and clinicians, greatly improving the feedback efficiency of evaluation and analysis and control effects, and enhancing the pertinence and effectiveness of control. It is a new type of intelligent neural control method.

Claims

1. A closed-loop adaptive transcranial electrical stimulation method, comprising a closed-loop adaptive transcranial electrical stimulation device comprising a transcranial electrical stimulation unit and an electroencephalogram (EEG) acquisition unit controlled by a control unit, wherein the transcranial electrical stimulation unit and the EEG acquisition unit share a common set of electrodes via a switching circuit controlled by the control unit; the transcranial electrical stimulation unit comprises positive and negative high-voltage circuits connected to the electrodes via a switching circuit, and the EEG acquisition unit comprises a filter circuit and an ADC acquisition circuit connected thereto, wherein the filter circuit is connected to the electrodes via a switching circuit. The method comprises the following steps: S1: Choose different experimental paradigms according to different subjects; S2: preset control target, which serves as the level setting for evaluating the brain state of the subject after control; S3: Preset the control model and determine the output parameters of transcranial electrical stimulation; S4: Collect and process EEG data to obtain the EEG state evaluation results of the subjects; S5: collect behavioral data and conduct paradigm evaluation; S6: Generate a fusion brain state evaluation algorithm based on the EEG precision evaluation algorithm and paradigm evaluation algorithm, and calculate the subject's state evaluation results; S7: Calculate the deviation between the subject's state evaluation result and the preset control target; S8: calling the control model to adjust the transcranial electrical stimulation output parameters according to the deviation value; S9: Repeat steps S3-S8, and circulate the online evaluation test control until there is no deviation between the "subject status evaluation result and the preset control target".

2. The closed-loop adaptive transcranial electrical stimulation method according to claim 1, wherein: The EEG acquisition unit includes an EEG acquisition circuit connected to electrodes, and can be constructed into 16 channels, 32 channels, 64 channels, and 128 channels by adopting a daisy-chain cascade.

3. The closed-loop adaptive transcranial electrical stimulation method according to claim 1, wherein: The transcranial electrical stimulation circuit is further provided with a constant current feedback detection circuit, which sends the stimulation current values ​​of the positive and negative high voltage circuits to the control unit.

4. The closed-loop adaptive transcranial electrical stimulation method according to claim 1, wherein: The switching circuit controls the connection between the electrodes and the transcranial electrical stimulation unit or the EEG acquisition unit. The control unit realizes the synchronous control mode and the asynchronous control mode of the electrodes by controlling the switching circuit. The synchronous control mode means that the electrodes perform transcranial electrical stimulation at the same time, or the electrodes perform EEG acquisition at the same time; the asynchronous control mode means that some electrodes perform EEG acquisition while some electrodes perform transcranial electrical stimulation.

5. The closed-loop adaptive transcranial electrical stimulation method according to claim 1, wherein: In step S3, the steps for setting the output parameters are as follows: S11: Determine the mode of transcranial electrical stimulation; It is divided into transcranial direct current stimulation, transcranial alternating current stimulation, transcranial pulse stimulation, and transcranial random stimulation modes; S12: Determine the location of transcranial electrical stimulation; The position is selected from the international standard EEG position distribution map, and there are 1 to N electrical stimulation positions, of which the positive pole is 1 to N-1 positions, and the negative pole is 1 to N-1 positions, which together constitute a control circuit combination, where N is a natural number greater than or equal to 2; S13: Determine the specific parameters of transcranial electrical stimulation; Including the polarity of the stimulation current, the duration of the stimulation current, the amplitude of the stimulation current, the bias of the stimulation current, the frequency of the stimulation current, and the duty cycle of the stimulation current; S14: Determine the evaluation method for regulatory objectives; There are two ways to evaluate control targets: asynchronous and synchronous. The asynchronous method evaluates whether the control target has been achieved after the control is completed, while the synchronous method evaluates whether the control target has been achieved while the control is being executed. S15: Determine the operating mode when the control target is achieved or not achieved; Stop the current regulation when the regulation goal is achieved, or continue regulation for a fixed period of time to consolidate the goal; When the control target is not achieved, the control can continue according to a set time, and the control can be stopped regardless of whether the control target is achieved; or the control can be stopped directly when the control target is not achieved.

6. The closed-loop adaptive transcranial electrical stimulation method according to claim 1, wherein: In step S4, the EEG state evaluation includes data preprocessing, relevant feature analysis and extraction, machine learning, and obtaining the EEG state evaluation results of the subject; The data preprocessing is intended to remove bad data or data full of artifacts without changing the clean data, or to change the clean data by applying filters or spatial transformations to make it conducive to analysis, including re-referencing, filtering, and independent principal component analysis. The related feature analysis and extraction is used to extract basic data for EEG analysis, including Fourier transform, nonlinear transformation, time domain analysis, and brain network analysis. The machine learning and obtaining of the subject's EEG state evaluation results realize the classification task in supervised learning, including the processing methods of support vector machine, neural network framework, and autoencoder network.

7. The closed-loop adaptive transcranial electrical stimulation method according to claim 1, wherein: In step S5, the paradigm behavior state evaluation includes: counting the subject's reaction time, counting the subject's reaction accuracy, and calculating the reaction time and reaction accuracy characteristics within and between subjects.

8. The closed-loop adaptive transcranial electrical stimulation method according to claim 1, wherein: In step S6, the accuracy reaction time of the deviation state and the non-deviation state is introduced into the discrimination model as a behavioral indicator. The first method is to perform feature fusion based on data of different modalities. The fusion methods include data-level fusion, decision-level fusion, and intermediate fusion. The second approach is to not consider accuracy response time as data of different modalities, but only as a general feature. Normalized with EEG features; The third method is to perform algorithm fusion of different data modalities based on the differences between the absolute control group populations.

9. The closed-loop adaptive transcranial electrical stimulation method according to claim 1, wherein: In step S5, paradigm evaluation includes but is not limited to the Posner attention paradigm and the n-back working memory paradigm.

Images