A neural stimulation system

Abstract

The application provides a nerve stimulation system, comprising: a signal generation module, a control module, a signal acquisition module, and a multiplexed electrode; the control module can control the signal generation module to generate a stimulation signal and output the stimulation signal to a target nerve through the multiplexed electrode; the control module can also control the signal acquisition module to acquire a potential signal of the target nerve through the multiplexed electrode; the control module processes the acquired potential signal as follows: inputting the acquired potential signal into a pre-trained evoked potential extraction model to obtain an evoked potential signal corresponding to the stimulation signal; and the control module adjusts the parameters of a subsequent stimulation signal based on the evoked potential signal. The application extracts the evoked potential signal from the potential signal acquired by the signal acquisition module through the pre-trained model, reduces the hardware requirements and energy consumption requirements of the signal acquisition module, and thus can provide a nerve stimulation system with smaller implant volume and longer effective use time.

Description

Technical Field

[0001] This application relates to the field of medical devices, and more specifically, to a nerve stimulation system. Background Technology

[0002] Neurostimulation has been used in the diagnosis and treatment of various diseases, such as electrical stimulation of peripheral or central nervous structures like the vagus nerve, sacral nerve, dorsal root ganglia of the spinal cord, trigeminal nerve, and glossopharyngeal nerve, to treat neurological diseases such as epilepsy, depression, Parkinson's disease, migraine, chronic inflammation, obesity, hypertension, and heart failure. However, several significant drawbacks exist in clinical applications: 1. Difficulty in acquiring evoked potentials for rapid feedback: Although theoretically evoked potentials can serve as objective indicators of neural responses, their amplitude is usually only one percent or even lower than the stimulation signal, and is severely interfered with by stimulation wake and electrode polarization effects. Especially in implantable devices, limited by size, power consumption, and analog front-end noise, traditional amplifiers struggle to acquire clear evoked potential signals immediately after stimulation; 2. Inconsistency between energy limitations and acquisition frequency: Implantable devices typically rely on batteries or wireless power, with extremely limited energy budgets. While high-frequency acquisition of evoked potentials helps improve feedback accuracy, it significantly shortens battery life or exceeds the safe power threshold for wireless power. To address one or more of these problems, this application proposes a neurostimulation system. Summary of the Invention

[0003] This application provides a neural stimulation system, comprising:

[0004] Signal generation module, control module, signal acquisition module, multiplexed electrodes;

[0005] The control module is configured to control the signal generation module to generate a stimulation signal and output it to the target nerve through the multiplexed electrode, and to control the signal acquisition module to acquire the potential signal generated by the target nerve after receiving the stimulation signal through the multiplexed electrode.

[0006] The control module is configured to process the acquired potential signals as follows:

[0007] The collected potential signals are input into a pre-trained evoked potential extraction model to obtain the evoked potential signals corresponding to the stimulus signals.

[0008] The control module is configured to adjust the parameters of subsequent stimulus signals based on the evoked potential signals.

[0009] Furthermore, the evoked potential extraction model loaded in the control module is obtained in the following manner:

[0010] After applying the stimulation signal, the signal acquisition module is used to collect potential signals from the target nerve as noisy sample signals, and the signal acquisition device is used to collect corresponding evoked potential signals from the target nerve as high-precision sample signals.

[0011] A training sample is formed by using the noisy sample signal as the sample to be labeled and the high-precision sample signal as the label.

[0012] The above steps are followed to generate a training sample set, which is used to train the initial deep learning model and obtain the evoked potential extraction model.

[0013] Optionally, the evoked potential extraction model loaded in the control module is fine-tuned before use. The fine-tuning process is as follows:

[0014] During the implantation of the neurostimulation system, the reusable electrode is connected to the target nerve, and then the electrode of the signal acquisition device is connected to the same target nerve.

[0015] A test signal is applied, and the signal acquisition module is used to collect potential signals from the target nerve as personalized noisy sample signals. At the same time, the signal acquisition device is used to collect corresponding evoked potential signals from the target nerve as personalized high-precision sample signals.

[0016] Using the personalized noisy sample signal as the sample to be labeled, and the personalized high-precision sample signal as the label, fine-tuned training samples are formed.

[0017] The evoked potential extraction model was trained and fine-tuned using the fine-tuned training samples.

[0018] Optionally, the initial deep learning model is a long short-term memory network.

[0019] Optionally, the signal acquisition device includes a signal amplification circuit, a subtraction circuit, and a filtering circuit. The subtraction circuit can subtract the baseline signal, and the filtering circuit can filter out noise signals. The power consumption of the signal acquisition module is lower than that of the signal acquisition device. In this application, the signal acquisition device can be set up independently or can be part of a system.

[0020] Optionally, the signal acquisition module further includes an analog-to-digital conversion unit, which is used to perform analog-to-digital conversion on the acquired potential signal.

[0021] Optionally, the neurostimulation system further includes a wireless communication module, through which the control module establishes a wireless communication connection with an external repeater or a programmable controller to transmit energy and / or information.

[0022] Optionally, the neural stimulation system further includes a power source that supplies power to the signal generation module, control module, and signal acquisition module. The power source may be a battery and / or an energy transfer module.

[0023] Optionally, the processing module further controls the working state of the signal generation module based on the extracted evoked potential signal, and adjusts the parameters of the stimulation signal.

[0024] The present invention also provides a peripheral nerve stimulation system, including any of the preceding nerve stimulation systems, for performing peripheral nerve stimulation.

[0025] The exemplary embodiments of this application can have the following beneficial effects:

[0026] 1. By using a pre-trained evoked potential extraction model, the potential signals acquired by the signal acquisition module can be processed to extract evoked potential signals, which reduces the hardware and energy consumption requirements of the signal acquisition module, thereby enabling a neurostimulation system with a smaller implantation volume and a longer effective usage time.

[0027] 2. During the collection of training samples, after applying the stimulation signal, the potential signals of the same target nerve are collected through the signal acquisition module and the signal acquisition device. The noisy sample signals collected by the signal acquisition module are directly used as the samples to be labeled, and the high-precision sample signals collected by the signal acquisition device are used as labels to form training samples. Based on the design of the same stimulation site and the same stimulation signal, the differences caused by the stimulation signal are eliminated, so that the trained evoked potential extraction model can directly extract the evoked potential signal from the collected noisy sample signal.

[0028] 3. During the implantation of the neurostimulation system, patient data was further collected. The model parameters were then fine-tuned and optimized using the data from the patients, making the extracted evoked potential signals closer to the patients' actual condition.

[0029] 4. The signal acquisition device includes a signal amplification circuit, a subtraction circuit, and a filtering circuit, which can remove baseline signals and noise signals.

[0030] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0031] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0032] Figure 1 A schematic diagram of the structure of the nerve stimulation system provided by the present invention is shown;

[0033] Figure 2 One of the schematic diagrams of the neural stimulation system used in the process of collecting training samples is shown;

[0034] Figure 3 The second schematic diagram shows the structure of the neural stimulation system used in the process of collecting training samples;

[0035] Figure 4 The diagram illustrates the process by which a long short-term memory network model operating in a neural stimulation system processes noisy sample signals and outputs evoked potential signals.

[0036] Figure 5 A schematic diagram of the stimulus signal output by the signal generation module is shown. Detailed Implementation

[0037] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar structures, and therefore their detailed descriptions will be omitted. Furthermore, the drawings are merely illustrative of this application and are not necessarily drawn to scale.

[0038] Although relative terms such as "upper" and "lower" are used in this specification to describe the relative relationship of one component of an icon to another, these terms are used only for convenience, such as according to the orientation of the examples in the accompanying drawings. It is understood that if the device of the icon is flipped so that it is upside down, the component described as "upper" will become the component described as "lower." When a structure is "upper" of another structure, it may mean that the structure is integrally formed on the other structure, or that the structure is "directly" mounted on the other structure, or that the structure is "indirectly" mounted on the other structure through another structure.

[0039] The terms “a,” “one,” “the,” and “at least one” are used to indicate the presence of one or more elements / components / etc.; the terms “include” and “have” are used to indicate an open-ended inclusion meaning and that there may be other elements / components / etc. in addition to the listed elements / components / etc.

[0040] Example 1

[0041] Reference Figure 1This embodiment provides a nerve stimulation system, including: a signal generation module, a control module, a signal acquisition module, and multiplexed electrodes;

[0042] The control module can control the signal generation module to generate stimulation signals and output them to the target nerve through multiplexed electrodes. The control signal acquisition module can acquire the potential signals generated by the target nerve after receiving the stimulation signals through multiplexed electrodes.

[0043] The control module processes the acquired potential signal as follows:

[0044] The collected potential signals are input into a pre-trained evoked potential extraction model to obtain the evoked potential signals corresponding to the stimulus signals.

[0045] The control module adjusts the parameters of subsequent stimulus signals based on the evoked potential signals.

[0046] Specifically, the two signal output terminals of the signal generation circuit are connected to multiplexed electrodes, which are in contact with the target nerve, forming a circuit for transmitting stimulation signals to the target nerve. The control module can control whether the signal generation circuit outputs stimulation signals to the multiplexed electrodes. For example, a switching transistor (such as a bipolar junction transistor or a field-effect transistor) is provided at at least one signal output terminal of the signal generation circuit. When a stimulation signal needs to be output, the control module adjusts the switching transistor to the on state; when a stimulation signal does not need to be output, the control module adjusts the switching transistor to the off state, cutting off the circuit for transmitting stimulation signals.

[0047] The two signal input terminals of the signal acquisition circuit are connected to multiplexed electrodes, which are in contact with the target nerve, forming a circuit for acquiring the target nerve's potential signal. The control module can control whether to acquire the target nerve's potential signal. For example, a switching transistor is provided at at least one signal input terminal of the signal acquisition circuit. When signal acquisition is needed, the control module adjusts the switching transistor to the on state; when signal acquisition is not needed, the control module adjusts the switching transistor to the off state, cutting off the potential signal acquisition circuit.

[0048] Overall, reusable electrodes can both output stimulation signals to achieve nerve stimulation and collect potential signals to monitor the stimulation effect.

[0049] Furthermore, since evoked potential signals are typically weak, amplification is required after acquisition. However, power amplifier circuits consume a lot of power, and high-frequency acquisition of potential signals significantly shortens battery life. Therefore, to ensure battery life, the signal acquisition frequency needs to be limited, which reduces the acquisition effect of evoked potentials. Moreover, evoked potentials are easily interfered with by other noise signals. To address these shortcomings of the prior art, this embodiment pre-trains an evoked potential extraction model that can extract evoked potential signals from the acquired raw signals, eliminate noise interference, and accurately reconstruct the evoked potential signals. The evoked potential extraction model is trained on data collected from scenarios involving stimulation of the same location. For example, if a nerve stimulation system is used to stimulate the vagus nerve, the evoked potential extraction model is trained on raw potential data and evoked potential data collected from the vagus nerve stimulation scenario; similarly, if a nerve stimulation system is used to stimulate the tibial nerve, the evoked potential extraction model is trained on raw potential data and evoked potential data collected from the tibial nerve stimulation scenario.

[0050] Furthermore, the evoked potential extraction model loaded in the control module is obtained in the following way:

[0051] First, training data for the evoked potential extraction model is collected using animal experiments. Taking vagus nerve stimulation as an example, a neurostimulation system and signal acquisition device are implanted into the vagus nerve of experimental dogs / monkeys. To collect more realistic data, the neurostimulation system and signal acquisition device can also be implanted into the vagus nerve of clinical trial patients. In this case, the quality of the collected training data is higher, and the accuracy of the trained evoked potential extraction model is higher. (Refer to...) Figure 2 Two sets of signal acquisition hardware were used: a signal acquisition module and a signal acquisition device with a higher sampling frequency and better signal processing capabilities. After the stimulation signal was applied by the signal generation module, data was simultaneously acquired using both the signal acquisition module and the signal acquisition device. Due to limitations in battery power and implantation size, the sampling frequency and signal processing capabilities of the signal acquisition module were restricted, and it acquired noisy sample signals from the target nerve. The signal acquisition device was used during the acquisition of training data and was not specifically designed for human signal acquisition. It was not limited by battery power or implantation size and could acquire high-precision sample signals from the target nerve, which could be directly used as the evoked potential signal corresponding to the stimulation signal.

[0052] Figure 2 In the example, the signal acquisition device is independent of the neural stimulation system. The signal acquisition device is removed after the training data is collected. Figure 3 In the example, the signal acquisition device is integrated into the neurostimulation system. The signal acquisition module and the signal acquisition device acquire potential signals in parallel. Because the two are integrated, implantation is more convenient. It is understandable that... Figure 3The current neural stimulation system is relatively large, but it can serve as a basic product for iterative updates. This basic product can be used for neural stimulation and also assist in the collection of training data.

[0053] During a single data acquisition process, the noisy sample signals acquired by the signal acquisition module can serve as samples to be labeled, while the high-precision sample signals acquired simultaneously by the signal acquisition device can serve as labels, forming a training sample. Following this data acquisition process, more training samples can be generated. For example, by changing the experimental subject, adjusting the stimulus signal, and re-acquiring data, more training samples can be generated. These training data constitute the training sample set.

[0054] Finally, the initial deep learning model is trained using the training sample set generated above. The model parameters are trained and optimized until the training objective is achieved (e.g., the model accuracy reaches the expected level, or the model error no longer decreases), and finally the evoked potential extraction model is obtained.

[0055] Furthermore, it should be noted that existing technologies for evoked potential extraction typically consider the influence of the stimulus signal (amplitude, frequency) on the evoked potential. In this embodiment, the signal acquisition module and signal acquisition device can acquire potential signals in parallel on the same target nerve. After applying the stimulus signal, the signal acquisition module and signal acquisition device acquire noisy sample signals and high-precision sample signals on the target nerve, respectively. The noisy sample signals and high-precision sample signals have the same acquisition location and the same stimulus signal. This embodiment directly associates the two to train the evoked potential extraction model. The evoked potential extraction model can directly extract the evoked potential signal from the data acquired by the signal acquisition module, resulting in higher data processing efficiency.

[0056] Example 2

[0057] Based on Example 1, the evoked potential extraction model loaded in the control module is further fine-tuned before use. The specific fine-tuning process is as follows:

[0058] For a specific patient, after placing the signal acquisition module onto the target nerve, the signal acquisition device is also placed in parallel onto the target nerve to jointly acquire the potential signals of the target nerve. It is understood that, similar to Example 2, the signal acquisition module and the signal acquisition device have different signal acquisition and processing capabilities, resulting in different signal quality.

[0059] First, fine-tuning training samples are collected: a test signal is applied, and a noisy sample signal is collected from the target nerve using a signal acquisition module. Simultaneously, a high-precision sample signal is collected from the target nerve using a signal acquisition device; this signal can be directly used as the corresponding evoked potential signal. The noisy sample signal and the high-precision sample signal are neural potential signals generated from the same test signal. The collected noisy sample signal is used as the sample to be labeled, and the high-precision sample signal is used as the label, forming a fine-tuning training sample. Following the above steps, more fine-tuning training samples can be generated, for example, by adjusting the amplitude and frequency of the test signal and re-acquiring the potential signals to generate new fine-tuning training samples.

[0060] The aforementioned fine-tuning training samples collected from patients form a fine-tuning training sample set, which is used to train and fine-tune the evoked potential extraction model. By utilizing data from patients, the model parameters are specifically fine-tuned and optimized, making the extracted evoked potential signals more closely resemble the patient's individual reality. Understandably, after collecting the fine-tuning training samples, the signal acquisition device is removed, leaving only the implantable signal acquisition module. This completes the collection of fine-tuning training data and the implantation of the neural stimulation system.

[0061] Example 3

[0062] Building upon Example 1, the initial deep learning model used is a Long Short-Term Memory (LSTM) network.

[0063] Specifically, the potential signals acquired by the aforementioned signal acquisition module and signal acquisition device are sequential signals over a period of time. Long Short-Term Memory (LSTM) networks have good processing capabilities for sequential signals. They can learn the temporal dependence of input data, distinguish the dynamic characteristics of evoked potential signals from noise signals, and better extract evoked potential signals.

[0064] Reference Figure 4 Long Short-Term Memory (LSTM) networks utilize three gated units—the forget gate, input gate, and output gate—to selectively store and transfer information. The forget gate, F... t Decide which old information should be retrieved from memory cell C t Delete; Input Gate I t Filter new input information and decide which new information should be added to the paired memory unit C. t Middle; Output Gate O t Based on the current cell state, generate output and determine which information in the memory unit should be output to the hidden state H at the current time step. t In the training and actual use of the model, the noisy sample signal sequence Xi is input sequentially to obtain the extracted signal sequence H. t That is, the evoked potential signal predicted by the model.

[0065] Example 4

[0066] Based on Example 1, the signal acquisition device further includes a signal amplification circuit, a subtraction circuit, and a filtering circuit. The subtraction circuit can subtract the baseline signal, and the filtering circuit can filter out noise signals.

[0067] It is understandable that signal acquisition devices support higher sampling frequencies, their signal amplification circuits operate for longer periods, and they also perform baseline signal filtering operations. Therefore, the power consumption of signal acquisition devices is higher than that of signal acquisition modules.

[0068] Furthermore, the raw potential signal acquired by the electrodes is sequentially passed through the amplification circuit, filtering circuit, and subtraction circuit of the signal acquisition device before being input to the control module, which improves the quality of the output signal. Even further, the signal acquisition device also includes another amplification circuit to amplify the signal before it is output to the control module.

[0069] Example 5

[0070] Based on the aforementioned embodiments, the signal acquisition module also includes an analog-to-digital conversion unit. The analog signal after preliminary processing (amplification and filtering) by the analog-to-digital conversion unit is converted into a digital signal to facilitate the next step of processing (inputting it into the potential extraction model to extract the potential signal).

[0071] Example 6

[0072] Based on the aforementioned embodiments, the nerve stimulation system further includes a wireless communication module, through which the control module establishes a wireless communication connection with an external repeater or a programmable controller to transmit energy and / or information.

[0073] Specifically, the neurostimulation system is an implantable device, and the external repeater can be a wearable device that can wirelessly charge the neurostimulation system and wirelessly receive and temporarily store / forward the evoked potential signals collected by the neurostimulation system, for example, forwarding them to a programmer. Optionally, the neurostimulation system can also directly communicate wirelessly with the external programmer, which can further analyze and process the evoked potential signals.

[0074] Furthermore, the programmable controller can set parameters for the neural stimulation system and repeaters, such as setting the amplitude of the stimulation signal, the duration of a single pulse, the duty cycle of continuous pulses, the period of continuous pulse sequences, the sampling frequency, and the charging power of the repeaters, etc.

[0075] Example 7

[0076] Based on the foregoing embodiments, the neural stimulation system also includes a power supply, which powers the signal generation module, control module, and signal acquisition module. The power supply is a battery and / or an energy transmission module.

[0077] Example 8

[0078] Based on the aforementioned embodiments, the processing module also controls the working state of the signal generation module according to the extracted evoked potential signal, and adjusts the parameters of the stimulation signal.

[0079] For example, when the evoked potential is too weak, it is considered that the stimulation effect may not be achieved, so the signal generation module is controlled to enhance the stimulation signal (e.g., increase the amplitude of the stimulation signal, increase the duration of a single pulse, or increase the duty cycle of continuous pulses). When the evoked potential is too strong, it is considered that it may cause discomfort or harm to the patient, so the signal generation module is controlled to weaken the stimulation signal (e.g., reduce the amplitude of the stimulation signal, reduce the duration of a single pulse, or reduce the duty cycle of continuous pulses).

[0080] Example 9

[0081] Based on the aforementioned embodiments, the stimulation signal output by the signal generation module includes a positive signal and a negative signal, such that the absolute value of the signal amplitude integral over time is the same, referring to... Figure 5 That is, the area of ​​S1 is equal to the area of ​​S2. It can be understood that the absolute values ​​of the integrals of the positive and negative signals are equal, which does not restrict the amplitude, pulse width, waveform, or number of pulses. For example, the number of pulses of the positive signal can be different from the number of pulses of the negative signal.

[0082] Prolonged stimulation can lead to charge accumulation at the electrode-tissue interface, forming a polarization voltage that further masks evoked potential signals and may even cause tissue damage or electrode corrosion. In this embodiment, the application of both positive and negative signals reduces charge accumulation, shortening the duration and amplitude of the tailing phenomenon.

[0083] Furthermore, within a stimulation cycle, the positive and negative signals are repeated several times; for example, a stimulation cycle may include four positive signals and four negative signals.

[0084] This embodiment significantly reduces charge accumulation and shortens the duration of the long-tail effect of the stimulus signal by using positive and negative stimulation signals. By setting the time interval between the evoked potential and the stimulus signal in a stimulation cycle, the influence of the long-tail effect of the stimulus signal on the evoked potential is eliminated. While ensuring the stimulation effect, timely feedback on the stimulus signal is obtained. The stimulus signal can be adjusted through the evoked potential to improve the effectiveness of the stimulation.

[0085] Example 10

[0086] Based on the aforementioned embodiments, this embodiment also provides a short-circuit branch between the two signal output terminals of the signal acquisition module, and a controlled short-circuit switch (such as a switching transistor) is provided on the short-circuit branch.

[0087] The control module is configured to turn on the controlled switch on the short-circuit branch before outputting the stimulation signal through the multiplexed electrode, thereby connecting the two multiplexed electrodes and eliminating the charge that has accumulated on the multiplexed electrode.

[0088] This embodiment eliminates accumulated charge through short-circuit connection, which can further improve the quality of the acquired signal.

[0089] Example 11

[0090] Building upon the aforementioned embodiments, the ratio of signal acquisition frequency to stimulation frequency in the signal acquisition module is further ensured to be within a preset range. Preferably, the preset ratio is not less than 1:1000, more preferably not less than 1:100, to ensure that sampling is not too sparse. Further, the preset ratio does not exceed 1:5, more preferably not more preferably 1:8, to avoid frequent signal acquisition which significantly shortens battery life (signal amplification is required during signal acquisition, a process with high energy consumption). Reducing the number of signal acquisitions can effectively reduce energy consumption, extend battery life, or reduce the power of wireless energy transmission. For example, in a wireless transmission system, the implanted part can be equipped with an energy storage structure (capacitor, etc.). During each stimulation signal transmission, the implanted part receives and stores a portion of the energy, waiting for sufficient stored energy to perform a signal acquisition.

[0091] This embodiment collects induced potentials at frequencies within a certain range, reducing energy consumption, extending battery life, or meeting power limits for safe energy transmission; the feedback collection ratio can also be adjusted in a personalized manner using historical data.

[0092] Example 12

[0093] This embodiment provides a peripheral nerve stimulation system, including the nerve stimulation system provided in any of the foregoing embodiments, for implantation into a patient's peripheral nerves to perform peripheral nerve stimulation. Examples of peripheral nerves include the vagus nerve, tibial nerve, and spinal nerves.

[0094] Other embodiments of this application will readily conceive of by those skilled in the art upon consideration of the specification and practice of the embodiments thereof. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not claimed in this application. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this application are indicated by the appended claims.

Claims

1. A neural stimulation system, characterized in that, include: Signal generation module, control module, signal acquisition module, multiplexed electrodes; The control module is configured to control the signal generation module to generate a stimulation signal and output it to the target nerve through the multiplexed electrode, and to control the signal acquisition module to acquire the potential signal generated by the target nerve after receiving the stimulation signal through the multiplexed electrode; The control module is configured to process the acquired potential signals as follows: The collected potential signals are input into a pre-trained evoked potential extraction model to obtain the evoked potential signals corresponding to the stimulus signals. The control module is configured to adjust the parameters of subsequent stimulation signals based on the evoked potential signals; The evoked potential extraction model loaded in the control module is obtained in the following way: A stimulation signal is applied, and the signal acquisition module is used to collect potential signals from the target nerve as noisy sample signals. At the same time, the signal acquisition device is used to collect corresponding evoked potential signals from the target nerve as high-precision sample signals. A training sample is formed by using the noisy sample signal as the sample to be labeled and the high-precision sample signal as the label. The above steps are followed to generate a training sample set, which is used to train the initial deep learning model and obtain the evoked potential extraction model.

2. A neural stimulation system according to claim 1, characterized in that, The evoked potential extraction model loaded in the control module was fine-tuned before use. The fine-tuning process is as follows: During the implantation of the neurostimulation system, the reusable electrode is connected to the target nerve, and then the electrode of the signal acquisition device is connected to the same target nerve. A test signal is applied, and the signal acquisition module is used to collect potential signals from the target nerve as personalized noisy sample signals. At the same time, the signal acquisition device is used to collect corresponding evoked potential signals from the target nerve as personalized high-precision sample signals. Using the personalized noisy sample signal as the sample to be labeled, and the personalized high-precision sample signal as the label, fine-tuned training samples are formed. The evoked potential extraction model was trained and fine-tuned using the fine-tuned training samples.

3. A neural stimulation system according to claim 1, characterized in that, The initial deep learning model is a long short-term memory network.

4. A neural stimulation system according to claim 1, characterized in that, The signal acquisition device includes a signal amplification circuit, a subtraction circuit, and a filtering circuit. The subtraction circuit can subtract the baseline signal, and the filtering circuit can filter out noise signals. The power consumption of the signal acquisition module is lower than that of the signal acquisition device.

5. A neural stimulation system according to claim 1, characterized in that, The signal acquisition module also includes an analog-to-digital conversion unit, which is used to perform analog-to-digital conversion on the acquired potential signal.

6. A neural stimulation system according to claim 1, characterized in that, The neurostimulation system also includes a wireless communication module, through which the control module establishes a wireless communication connection with an external repeater or programmable controller to transmit energy and / or information.

7. A neural stimulation system according to claim 1, characterized in that, The neurostimulation system also includes a power source, which supplies power to the signal generation module, control module, and signal acquisition module. The power source is a battery and / or an energy transmission module.

8. A neural stimulation system according to claim 1, characterized in that, The control module also controls the working state of the signal generation module based on the extracted evoked potential signal, and adjusts the parameters of the stimulation signal.

9. A peripheral nerve stimulation system, characterized in that, Includes the neural stimulation system according to any one of claims 1-8, for performing peripheral nerve stimulation.

Images