Intelligent closed-loop vagus nerve stimulation regulation method and system

Abstract

The present application relates to the technical field of medical equipment, and more particularly to a method and system for intelligent closed-loop vagus nerve stimulation control. The method comprises the following steps: collecting electroencephalogram signals and stimulation signals in real time, and obtaining future N-step electroencephalogram signal prediction values through a dynamic electroencephalogram signal prediction model; calculating a depression biomarker estimate value for representing the brain activity state at the next time point according to the future N-step electroencephalogram signal prediction values; calculating the optimal stimulation signal at the next time according to the difference between the depression biomarker target value and the depression biomarker estimate value; updating the stimulation signal to stimulate the vagus nerve according to the optimal stimulation signal, and updating the stimulation signal. The system for executing the above method is also disclosed. The optimal stimulation parameter is calculated according to the difference between the predicted brain activity state at the next time and the target state, so as to control the neural activity and improve the adjustment efficiency and control accuracy of the transcutaneous vagus nerve stimulation system.

Description

Technical Field

[0001] This invention relates to the field of medical device technology, and in particular to an intelligent closed-loop vagus nerve stimulation and control method and system. Background Technology

[0002] Currently, the treatment of depression primarily relies on medication, combined with psychotherapy, physical therapy, and other comprehensive treatments. Vagus nerve stimulation (VNS) modulates brain function by electrically stimulating the vagus nerve and is currently used for neurostimulation therapy in severe treatment-resistant depression (TRD). This technology uses implantable or non-implantable devices to send electrical signals to the vagus nerve, modulating brain regions related to mood processing (such as the amygdala and prefrontal cortex), improving mood and attention in patients with depression. Early clinical trials have shown effective and durable evidence of VNS's antidepressant efficacy in treating TRD. VNS has also been found to help improve the quality of life and suicide rate in patients with unipolar TRD, and to improve depression in patients with bipolar TRD.

[0003] A patent with publication number CN117899356B discloses an adaptive auricular vagus nerve stimulation device. The device includes an EEG acquisition module that collects real-time EEG signals, an ECG acquisition module that collects real-time heart rate variability signals, and a data processing module that analyzes these signals within a preset time period based on a pre-defined correspondence to obtain an adaptive stimulation current value. The auricular vagus nerve stimulation module applies a target stimulation current to the patient's auricular vagus nerve and adjusts the applied target stimulation current at preset time intervals based on the adaptive stimulation current value. This achieves adaptive adjustment of the closed-loop stimulation current for the vagus nerve, improving the sensitivity and accuracy of the adaptive adjustment. However, existing technologies only control the amplitude of the current and do not consider the frequency. Furthermore, while current adjustment is based on the difference between the current and target EEG and ECG signals, the EEG signals are constantly changing during actual stimulation, leading to low accuracy when adjusting the current based on the current EEG signal. This makes it impossible to achieve optimal stimulation in real-time. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent closed-loop vagus nerve stimulation and modulation method and system to solve the above-mentioned technical problems.

[0005] To achieve the above objectives, the present invention provides an intelligent closed-loop vagus nerve stimulation modulation method, the specific steps of which are as follows:

[0006] Step S1: Collect resting-state EEG signals before stimulation, and confirm the target brain activity state based on the resting-state EEG signals. Obtain the target value of the biomarker for depression based on the target brain activity state.

[0007] Step S2: Acquire EEG signals and stimulation signals during real-time stimulation, preprocess and decompose the EEG signals acquired during stimulation, extract power features and standardize the power features;

[0008] Step S3: Input the collected stimulus signals and standardized power features into the dynamic EEG signal prediction model to obtain the predicted EEG signal values ​​for the next N steps;

[0009] Step S4: Calculate the estimated values ​​of depression biomarkers to characterize the brain activity state at the next moment based on the predicted values ​​of EEG signals for the next N steps;

[0010] Step S5: Calculate the optimal stimulus signal for the next time step based on the difference between the target value and the estimated value of the depression biomarker, and update the stimulus signal according to the optimal stimulus signal to stimulate the vagus nerve;

[0011] Step S6: Update the stimulus signal and repeat steps S2-S5.

[0012] Preferably, in step S2, the preprocessing includes filtering and artifact removal; extracting power features of different frequency bands of the EEG signal and calculating the average power within the overlapping time window, removing the mean and dividing by the standard deviation to obtain the standardized power features.

[0013] The preferred training process for the dynamic EEG signal prediction model is as follows:

[0014] Dataset Acquisition: Brain networks were stimulated with stimulation waves containing amplitude and frequency data, using parameterized randomized waveforms of equal power across the entire frequency band. Obtain full-band dynamic EEG signal Y of brain network t For each trial, all input stimuli U t ={u1,u2,...,u t} and output signal Y t ={y1,y2,...,y t The dataset {u} is composed of t ,y t};

[0015] Dataset partitioning: The above dataset was partitioned into training and test sets using four-fold cross-validation.

[0016] A dynamic EEG signal prediction model was constructed, and the dynamic EEG signal prediction model is as follows:

[0017]

[0018] Among them, y t Let x be the EEG signal of the t-th time segment.t To use artificial neural networks to obtain y t The latent states extracted from them, x is a set of potential representations of brain functional network states. t+1 To use artificial neural networks to obtain y t+1 The latent states are extracted from the matrix; A, B, and C are the hidden state feature matrix, input matrix, and output matrix, respectively.

[0019] Training and testing: The constructed dynamic EEG signal prediction model is trained on the training set until it converges or reaches the required number of iterations, and then tested on the test set until the accuracy evaluation index of the dynamic EEG signal prediction model reaches the set value.

[0020] Preferably, the estimated values ​​of depression biomarkers are calculated based on the predicted values ​​of EEG signals in the next N steps. The estimated values ​​of depression biomarkers include the individual's alpha band peak frequency and relative γ band power.

[0021] Preferably, step S5 specifically includes:

[0022] Construct a cost function based on the target brain activity state, the cost function j is as follows:

[0023]

[0024] Among them, s t Here are the estimated values ​​of depression biomarkers at time t. For s t transpose, s t+N This is the estimated value of the biomarker for depression predicted in step N. For s t+N The transpose of , Q is the state weight matrix, which adjusts the importance of different state variables, R is the control input weight matrix, which adjusts the penalty for the magnitude of the control input, and F is the terminal state weight matrix, which is used to ensure the stability of the system at the end of the prediction period.

[0025] Solving for minJ within the constraints of the safety stimulus parameters, where min is the minimum value, yields the finite-time control sequence u. t+1 ~u t+N-1 , take u t+1 The optimal parameters for the input stimulus in the next time period

[0026] Preferably, the optimal parameters Merged into U t Get U t+1 The electroencephalogram (EEG) signal y was measured after stimulation. t+1 Merged into Y t Get Y t+1The dynamic EEG signal prediction model was refitted using the new dataset to update the predicted EEG signals for the next N steps.

[0027] A system for implementing an intelligent closed-loop vagus nerve stimulation modulation method includes:

[0028] The electrical stimulation module is used to output a stimulation current based on the stimulation signal. The stimulation parameters of the stimulation current include the current intensity and the oscillation frequency.

[0029] EEG signal acquisition module, several EEG signal acquisition modules are integrated into a wearable device, and several EEG signal acquisition modules correspond to closed loop vagus nerves at different locations. Several EEG signal acquisition modules are connected to multiplexers or several analog-to-digital converters with amplification functions.

[0030] The regulation and calculation module is used for signal processing, constructing and training dynamic EEG signal prediction models, and calculating optimal stimulus signals.

[0031] The signal transmission module is used to enable communication between the EEG signal acquisition module and the modulation calculation module, as well as between the electrical stimulation module and the modulation calculation module.

[0032] The coordination module, electrical stimulation module, EEG signal acquisition module, modulation calculation module, and signal transmission module are all electrically connected to the coordination module to achieve coordinated control and event recording of the electrical stimulation module, EEG signal acquisition module, modulation calculation module, and signal transmission module.

[0033] Therefore, the present invention employs the above-mentioned intelligent closed-loop vagus nerve stimulation modulation method and system, which has the following beneficial effects:

[0034] (1) The collected stimulus signals are input into the dynamic EEG signal prediction model to obtain the predicted EEG signal values ​​for the next N steps. Based on the predicted EEG signal values ​​for the next N steps, the estimated values ​​of the depression biomarkers used to characterize the brain activity state at the next moment are calculated. Based on the difference between the target value of the depression biomarker and the estimated value of the depression biomarker, the optimal stimulus signal for the next moment is calculated, thus realizing dynamic adjustment of the stimulus signal with high accuracy.

[0035] (2) The control parameters in the stimulation signal include amplitude and frequency to achieve more precise functional adjustment.

[0036] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0037] Figure 1 This is a flowchart of an intelligent closed-loop vagus nerve stimulation and modulation method according to the present invention;

[0038] Figure 2This is a system block diagram of the present invention. Detailed Implementation

[0039] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use. They are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," and "connect" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0040] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0041] like Figure 1 As shown, a smart closed-loop vagus nerve stimulation modulation method includes the following specific steps:

[0042] Step S1: Collect resting-state EEG signals before stimulation, and confirm the target brain activity state based on the resting-state EEG signals. Obtain the target value of the biomarker for depression based on the target brain activity state.

[0043] Step S2: Acquire EEG signals and stimulation signals during real-time stimulation. Preprocess and frequency band decomposition are performed on the acquired EEG signals. Preprocessing includes filtering and artifact removal. Filtering reduces the EEG signals to the 1-45Hz frequency band. Stimulation artifacts are identified and removed to obtain clean EEG signal data.

[0044] Power features of different frequency bands in the EEG signal were extracted, and the average power within the overlapping time window was calculated, including different frequency bands such as 1-8Hz, 8-12Hz, 12-30Hz, and 30-45Hz. The length of the overlapping time window was 10 seconds, and the time step was 1 second. After removing the mean, the power features were divided by the standard deviation to obtain standardized power features suitable for subsequent analysis.

[0045] Step S3: Input the collected stimulus signals and standardized power features into the dynamic EEG signal prediction model to obtain the predicted EEG signal values ​​for the next N steps.

[0046] The training process of the dynamic EEG signal prediction model is as follows:

[0047] Dataset Acquisition: The brain network is stimulated with stimulation waves of randomized waveforms with equal power across the entire frequency band. In practice, a continuous stimulation method is used, randomly selecting a set of stimulation waveforms determined by amplitude and frequency parameters every 1 second for vagal nerve stimulation. The stimulation waves contain amplitude and frequency data. The amplitude and frequency combination of the electrical stimulation waveform includes:

[0048] {(0mA,0Hz),(150mA,5Hz),(150mA,20Hz),(280mA,5Hz),

[0049] (280mA, 20Hz)}, one of the following is randomly selected for each stimulus, and the probability of each combination being selected is as follows: Ten consecutive trials with identical stimulation protocols were conducted, each consisting of 180 stimulations lasting 1 second, for a total duration of 30 minutes. Dynamic electroencephalogram (EEG) signals of the full-band brain network were obtained. t For each trial, all input stimuli U t ={u1,u2,...,u t} and output signal Y t ={y1,y2,...,y t The dataset {u} is composed of t ,y t};

[0050] Dataset partitioning: The above dataset is divided into training and test sets using a four-fold cross-validation method. For each cross-validation, three-quarters of each experimental signal is used as the training set, and the two sets are concatenated. The remaining signals from each experiment are used as the test set.

[0051] A dynamic EEG signal prediction model was constructed, and the dynamic EEG signal prediction model is as follows:

[0052]

[0053] Among them, y t Let x be the EEG signal of the t-th time segment. t To use artificial neural networks to obtain y t The latent states extracted from them, x is a set of potential representations of brain functional network states. t+1 To use artificial neural networks to obtain y t+1 The latent states are extracted from the latent space; A, B, and C are the hidden state feature matrix, input matrix, and output matrix, respectively; the parameters of the hidden state feature matrix A, input matrix B, and output matrix C in the latent space of the brain network are estimated using the dynamic pattern decomposition method to characterize the overall dynamic state of the brain network.

[0054] Training and Testing: The constructed dynamic EEG signal prediction model is trained on the training set until convergence or the required number of iterations is reached. It is then tested on the test set until the accuracy evaluation metric of the dynamic EEG signal prediction model reaches a set value. The linear correlation coefficient (LCC) between the predicted brain network dynamics (i.e., the power characteristics of the EEG signal) and the actual signal power characteristics is calculated on the test set as a quantitative indicator of model accuracy. The formula for calculating the linear correlation coefficient is as follows:

[0055]

[0056] in, The power characteristic value of the EEG signal predicted by the model. represents the actual EEG signal power characteristic value, Cov() is the covariance function, and Var() is the variance function.

[0057] Step S4: Calculate estimated values ​​of biomarkers for depression to characterize brain activity at the next moment, based on predicted EEG signal values ​​for the next N steps. These estimated biomarkers include individual alpha peak frequency and relative gamma power. The baseline is defined as the estimated value of biomarkers representing a clinically normal state of brain network activity.

[0058] Step S5: Calculate the optimal stimulus signal for the next time step based on the difference between the target value and the estimated value of the depression biomarker.

[0059] A cost function J based on the target brain activity state is constructed as follows:

[0060]

[0061] Among them, s t Here are the estimated values ​​of depression biomarkers at time t. For s t transpose, s t+N This is the estimated value of the biomarker for depression predicted in step N. For s t+N The transpose of , Q is the state weight matrix, which adjusts the importance of different state variables, R is the control input weight matrix, which adjusts the penalty for the magnitude of the control input, and F is the terminal state weight matrix, which is used to ensure the stability of the system at the end of the prediction period.

[0062] Solving for minJ within the constraints of the safety stimulus parameters, where min is the minimum value, yields the finite-time control sequence u. t+1 ~u t+N-1 , take u t+1 The optimal parameters for the input stimulus in the next time period The vagus nerve is stimulated by updating the stimulation signal based on the optimal stimulation signal.

[0063] Step S6: Update the stimulus signal and repeat steps S2-S5.

[0064] Optimal parameters Merged into U t Get U t+1 The electroencephalogram (EEG) signal y was measured after stimulation. t+1 Merged into Y t Get Y t+1 The dynamic EEG signal prediction model was refitted using the new dataset to update the predicted EEG signals for the next N steps.

[0065] A system that performs the above-mentioned intelligent closed-loop vagus nerve stimulation modulation method, such as Figure 2 As shown, it includes:

[0066] The electrical stimulation module outputs a stimulation current based on a stimulation signal. It can apply various waveforms, including direct current (tDCS), oscillating current at a specified frequency (tACS), and pulsed or modulated current. The stimulation parameters include current intensity and oscillation frequency. The module features parameter adjustment capabilities, allowing for optimal adjustment of the stimulation current. It typically includes a controllable current source or voltage source. The design employs a bipolar constant current driver to generate current between any two electrodes, with channels switched by an analog switching network. The module incorporates electrode impedance detection; by injecting a small test current or measuring impedance using an existing stimulation current, if poor electrode contact (excessive impedance) is detected, the system will stop stimulation to prevent voltage overload.

[0067] The EEG signal acquisition module consists of a set of electrodes and related front-end electronic components. The electrodes are dry contact sensors, using conductive foam or semi-dry sponge material to improve contact issues, enhance convenience, increase user comfort, and speed up device installation. Several EEG signal acquisition modules are integrated into the wearable device, and these modules correspond to different locations of the closed-loop vagus nerve. This embodiment uses eight channels of EEG signals: Fp1 (left frontal pole), Fp2 (right frontal pole), F7 (left anterior temporal pole), F8 (right anterior temporal pole), T3 (left parietal lobe), T4 (right parietal lobe), P3 (left middle temporal pole), and P4 (right middle temporal pole). This can be expanded as needed; users can connect additional sensor modules to predetermined locations on the head-mounted device. The device will automatically identify the new electrodes and incorporate them into the EEG signal acquisition module array, thereby increasing the number of channels or covering additional areas. EEG signals from all active electrodes are output via a multiplexer or an analog-to-digital converter with amplification capabilities.

[0068] The control and calculation module includes a model prediction controller and an external display and operation device. The external display and operation device allows clinicians or users to view the system status and related parameter settings. The model prediction controller is used for signal processing, constructing and training a dynamic EEG signal prediction model, calculating the optimal stimulation signal, and transmitting the calculated optimal stimulation parameters to the electrical stimulation module through the signal transmission module to execute the adjusted stimulation.

[0069] The control and computing module can also connect to cloud services to enhance processing power and data storage, or it can run offline without relying on external services. This makes the entire closed-loop system portable and wearable, which has significant advantages over solutions that rely on laboratory equipment or magnetic resonance imaging.

[0070] The signal transmission module enables communication between the EEG signal acquisition module and the modulation calculation module, as well as between the electrical stimulation module and the modulation calculation module. Wireless transmission is achieved using Bluetooth Low Energy, Wi-Fi, 4G, or 5G modules. Alternatively, direct data transmission can be achieved via a wired interface, such as a USB interface. The transmission process includes error detection and encryption to ensure the privacy of the EEG data. The time from EEG acquisition to stimulation commands can be controlled within 100 milliseconds.

[0071] The coordination module, including a microcontroller or embedded processor, is electrically connected to the electrical stimulation module, the electroencephalogram (EEG) signal acquisition module, the modulation calculation module, and the signal transmission module, enabling coordinated control and event recording of these modules.

[0072] For example, a certain current limit (e.g., maximum 2mA) or a certain duration may not be allowed unless a safety authentication mode is explicitly activated.

[0073] For transcranial direct current stimulation, gradually increasing or decreasing the current can prevent user discomfort, and the current can be automatically and smoothly increased or decreased to avoid sudden changes.

[0074] If communication between the control and calculation modules is interrupted, the coordination module can immediately terminate any ongoing stimuli or revert to the default program.

[0075] The coordination module can also perform basic signal processing—for example, it may calculate coarse EEG indicators, such as whether the EEG signal is within the normal range, as a backup trigger mechanism to ensure that the user's brain activity remains within a safe range during closed-loop operation.

[0076] The coordination module includes a memory for recording events (when a stimulus was applied, parameters were set, etc.) for post-event analysis or regulatory compliance.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent closed-loop vagus nerve stimulation and modulation system, characterized in that, include: The electrical stimulation module is used to output a stimulation current based on the stimulation signal. The stimulation parameters of the stimulation current include the current intensity and the oscillation frequency. An electroencephalogram (EEG) signal acquisition module is integrated into a wearable device, with each EEG signal acquisition module corresponding to a closed-loop vagus nerve at a different location. The EEG signal acquisition modules are connected to a multiplexer or several analog-to-digital converters with amplification functions. The module is used to acquire resting-state EEG signals before stimulation and EEG signals during real-time stimulation. The regulation and calculation module is used for signal processing, constructing and training dynamic EEG signal prediction models, and calculating optimal stimulus signals. The dynamic EEG signal prediction model is as follows: in, The EEG signal for the t-th time segment is... To use artificial neural networks from The latent states extracted from them, , A set of potential representations of brain functional network states. To use artificial neural networks from The latent states are extracted from the matrix; A, B, and C are the hidden state feature matrix, input matrix, and output matrix, respectively. The optimal stimulus signal is calculated as follows: The target brain activity state is confirmed based on the resting-state electroencephalogram signal, and the target value of the biomarker for depression is obtained based on the target brain activity state. EEG signals and stimulation signals are acquired during real-time stimulation. The EEG signals acquired during stimulation are preprocessed and decomposed into frequency bands, and power features are extracted and standardized. The collected stimulus signals and standardized power features are input into the dynamic EEG signal prediction model to obtain the predicted EEG signal values ​​for the next N steps. Estimates of biomarkers for depression, used to characterize brain activity at the next moment, are calculated based on predicted EEG signal values ​​for the next N steps. The optimal stimulus signal for the next time step is calculated based on the difference between the target value and the estimated value of the depression biomarker. Construct a cost function based on the target brain activity state. as follows: in, Here are the estimated values ​​of depression biomarkers at time t. for transpose, This is the estimated value of the biomarker for depression predicted in step N. for The transpose of , Q is the state weight matrix, which adjusts the importance of different state variables, R is the control input weight matrix, which adjusts the penalty for the magnitude of the control input, and F is the terminal state weight matrix, which is used to ensure the stability of the system at the end of the prediction period. Let be the stimulus wave at time t. for Transpose of; Solve under constraints within the range of safe stimulus parameters. ,in, To find the minimum value, the control sequence in the finite time domain is obtained. ,Pick The optimal parameters for the input stimulus in the next time period ; The signal transmission module is used to enable communication between the EEG signal acquisition module and the modulation calculation module, as well as between the electrical stimulation module and the modulation calculation module. The coordination module, electrical stimulation module, EEG signal acquisition module, modulation calculation module, and signal transmission module are all electrically connected to the coordination module to achieve coordinated control and event recording of the electrical stimulation module, EEG signal acquisition module, modulation calculation module, and signal transmission module.

2. The intelligent closed-loop vagus nerve stimulation and modulation system according to claim 1, characterized in that: Signal processing includes filtering and artifact removal; power features of different frequency bands of EEG signals are extracted and the average power within the overlapping time window is calculated. The mean is removed and then divided by the standard deviation to obtain the standardized power features.

3. The intelligent closed-loop vagus nerve stimulation and modulation system according to claim 2, characterized in that, The training process of the dynamic EEG signal prediction model is as follows: Dataset Acquisition: Brain networks were stimulated with stimulation waves containing amplitude and frequency data, using parameterized randomized waveforms of equal power across the entire frequency band. A single-channel, full-band dynamic electroencephalogram (EEG) signal of the brain network was obtained. All input stimuli in each trial With single-channel output signal The dataset ; Dataset partitioning: The above dataset was partitioned into training and test sets using four-fold cross-validation. Constructing a dynamic electroencephalogram (EEG) signal prediction model; Training and testing: The constructed dynamic EEG signal prediction model is trained on the training set until it converges or reaches the required number of iterations, and then tested on the test set until the accuracy evaluation index of the dynamic EEG signal prediction model reaches the set value.

4. The intelligent closed-loop vagus nerve stimulation and modulation system according to claim 3, characterized in that: Estimated values ​​of depression biomarkers are calculated based on predicted EEG signals for the next N steps. These estimated values ​​include individual... Frequency band peak frequency, relative Frequency band power.

5. The intelligent closed-loop vagus nerve stimulation and modulation system according to claim 4, characterized in that, Optimal parameters Merging get Electroencephalogram (EEG) signals were measured after stimulation. Merging get The dynamic EEG signal prediction model was refitted using the new dataset to update the predicted EEG signals for the next N steps.

Images