Closed-loop electroacupuncture stimulation regulation and control method, system and device based on LSTM and medium

Abstract

The invention provides a closed-loop electro-acupuncture stimulation regulation and control method, system and device based on LSTM and a medium, and the method comprises the steps: obtaining physiological state data of a patient, and extracting time domain features and frequency domain features of the physiological state data; importing the time domain features and the frequency domain features into an LSTM model, and generating a state prediction score and a state prediction trend; and adjusting the pulse width according to the state prediction score, the electrical stimulation frequency and current intensity, and the state prediction trend. The problem that in the prior art, electroacupuncture stimulation parameters cannot be dynamically adjusted in real time according to the physiological state of a patient and can only be set according to fixation and experience, and consequently side effects are obvious is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical treatment, and particularly to a closed-loop electroacupuncture stimulation regulation method, system, device and medium based on LSTM. Background Art

[0002] In the prior art, although the closed-loop electroacupuncture stimulation system has made progress in multi-modal signal fusion (such as electroencephalogram, neurochemical sensing) and real-time monitoring, there are still the following key defects: 1. Uncertain treatment response: Traditional electroacupuncture stimulation parameters (such as current intensity, frequency) are mostly fixed or empirically set, lacking a dynamic adjustment mechanism based on the real-time physiological state of the patient. For example, although EEG signals can reflect brain activities, they cannot directly correlate with the degree of inflammation in abdominal infections, resulting in insufficient matching between stimulation parameters and the inflammatory state, and large fluctuations in treatment effects. 2. Obvious side effects: Static parameters are prone to cause over-stimulation or under-stimulation. For example, a fixed high current intensity may damage tissues or cause pain, while low-frequency stimulation may not effectively inhibit the release of inflammatory factors. 3. Lack of dynamic regulation ability: Existing systems rely on offline data analysis or manual intervention and cannot respond to changes in physiological parameters in real time. For example, the detection of inflammatory markers (such as IL-6, CRP) usually relies on laboratory analysis, with a delay of up to several hours, making it difficult to provide immediate feedback for electroacupuncture parameter adjustment. However, by combining heart rate variability (HRV) monitoring, intelligent LSTM models and closed-loop control technology, dynamic adjustment of electroacupuncture stimulation can be achieved, improving treatment effects and personalization levels. Summary of the Invention

[0003] Aiming at the deficiencies in the prior art, the present invention provides a closed-loop electroacupuncture stimulation regulation method, system, device and medium based on LSTM, which solves the problem in the prior art that electroacupuncture stimulation parameters cannot be dynamically adjusted in real time according to the physiological state of the patient and can only be set according to fixed and empirical values, resulting in obvious side effects.

[0004] According to an embodiment of the present invention, a closed-loop electroacupuncture stimulation regulation method based on LSTM includes:

[0005] Obtain the physiological state data of the patient, and extract the time-domain features and frequency-domain features of the physiological state data;

[0006] Import the time-domain features and frequency-domain features into the LSTM model to generate a state prediction score and a state prediction trend;

[0007] Adjust the frequency and current intensity of the electrical stimulation according to the state prediction score, and adjust the pulse width according to the state prediction trend.

[0008] Preferably, after obtaining the physiological state data, the physiological state data needs to be preprocessed, and then a sliding window is used to segment the physiological state data to obtain multiple data sequences, and time-domain features and frequency-domain features are extracted for each data sequence.

[0009] Preferably, the size of the sliding window is adjusted according to the prediction score obtained last time, and the sliding step of the sliding window is adjusted according to the state prediction trend obtained last time.

[0010] Preferably, the time-domain features include SDNN and RMSSD, and the frequency-domain features include LF / HF. Then, SDNN, RMSSD, and LF / HF are combined into a three-dimensional input sequence and imported into the LSTM model.

[0011] Preferably, the method for adjusting the frequency and current intensity of the electrical stimulation according to the state prediction score includes:

[0012] Setting a base frequency and a base current intensity according to the patient's physiological state;

[0013] Taking the difference between the state prediction score and the preset score threshold as the score increment, and then calculating the score change rate of the score increment per unit time;

[0014] Determining a gain coefficient according to the score change rate, and then increasing the base frequency and the base current intensity according to the gain coefficient and the score increment.

[0015] Preferably, the setting formula of the base current intensity is as follows:

[0016] I base = α·body weight + β·SDNN 初始 + γ

[0017] The setting formula of the base frequency is as follows:

[0018] f base = λ·age -1 + μ·RMSSD 初始

[0019] Wherein, the coefficients α, β, γ, λ, and μ are all constants, obtained by training with clinical data, and SDNN initial and RMSSD initial are the SDNN and RMSSD corresponding to the physiological state information of the patient collected for the first time.

[0020] On the other hand, according to an embodiment of the present invention, there is also provided a closed-loop electroacupuncture stimulation regulation system based on LSTM. This system uses the above-mentioned closed-loop electroacupuncture stimulation regulation method based on LSTM, and includes:

[0021] A monitoring module, which is used to collect the physiological state data of the patient;

[0022] An inference module, which is used to extract the time-domain features and frequency-domain features of physiological state data, obtain the corresponding state prediction score and state prediction trend according to the time-domain features and frequency-domain features, and then calculate the adjustment amplitudes of frequency, current intensity, and pulse width according to the state prediction score and state prediction trend;

[0023] An electroacupuncture stimulation module, which is used to adjust the electroacupuncture stimulation parameters according to the adjusted frequency, current intensity, and pulse width.

[0024] On the other hand, according to an embodiment of the present invention, there is also provided a computer, including at least one processor and a memory, where the memory stores a computer program, and the computer program is configured to be executed by the processor to implement the above-mentioned closed-loop electroacupuncture stimulation regulation method based on LSTM.

[0025] On the other hand, according to an embodiment of the present invention, there is also provided a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium, and the computer program can be executed by one or more processors to implement the above-mentioned closed-loop electroacupuncture stimulation regulation method based on LSTM.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] The present invention collects the physiological state data of the patient during the electroacupuncture stimulation process, extracts the time-domain features and frequency-domain features of the data, and uses the LSTM model to predict the physiological state score value and change trend in the short term, so as to adjust the frequency, current intensity, and pulse width of the electroacupuncture stimulation, so that the electroacupuncture stimulation is in a dynamically adjusted state throughout the treatment process and always provides appropriate electrical stimulation to the patient, without causing side effects such as overstimulation or understimulation. Description of the Drawings

[0028] Figure 1 It is a flowchart of the closed-loop electroacupuncture stimulation regulation method according to an embodiment of the present invention.

[0029] Figure 2 It is an architecture diagram of the closed-loop electroacupuncture stimulation regulation system according to an embodiment of the present invention. Detailed Embodiments

[0030] The technical solutions in the present invention will be further described below with reference to the drawings and embodiments.

[0031] As Figure 1 shown, an embodiment of the present invention proposes a closed-loop electroacupuncture stimulation regulation method based on LSTM, including:

[0032] Obtain the physiological state data of the patient, and extract the time-domain features and frequency-domain features of the physiological state data;

[0033] In the present invention, the heart rate variability (HRV) is used as the physiological state data of the collected patient. After obtaining the HRV data of the patient, the data signal within the time interval from R wave to R wave is intercepted from the HRV data, and then the intercepted HRV data is preprocessed:

[0034] (1) Denoising processing: The median filtering method is used to eliminate instantaneous noise, such as electromyogram interference.

[0035] (2) IQR outlier processing: Calculate the interquartile range (IQR) of the data to remove outliers caused by signal distortion or motion artifacts.

[0036] (3) Normalization: First, use Z-score standardization to eliminate the influence of dimensions, and then use Min-Max normalization to map the data to the interval [0,1]:

[0037]

[0038] After processing the data, the sliding window mechanism is used to divide the HRV data into multiple data sequences to increase the number of inputs of the subsequent LSTM model. In the present invention, when the HRV data of the patient is collected for the first time, it is intercepted with a sliding window size of 5 and a sliding step of 1. When intercepting the data sequence next time, the size of the sliding window is adjusted according to the prediction score obtained last time, and the sliding step of the sliding window is adjusted according to the state prediction trend obtained last time:

[0039]

[0040] step n =step n-1 ±1

[0041] where ΔS n-1 is the score increment of the previous state prediction score, l is the sliding window size, step is the sliding step. If the state prediction trend is remission, then -1; if the state prediction trend is deterioration, then +1.

[0042] After dividing the data sequences, for each data sequence, extract its time-domain features and frequency-domain features:

[0043] (1) Calculation of HRV time-domain features

[0044] SDNN (standard deviation): Reflects the overall heart rate variability, and the calculation formula is:

[0045]

[0046] RMSSD (Root Mean Square of the Successive Differences of RR Intervals): Measures vagal nerve activity, and the calculation formula is:

[0047]

[0048] (2) Calculation of HRV frequency domain features

[0049] LF / HF ratio: Calculates the ratio of low-frequency (LF, 0.04 - 0.15 Hz) and high-frequency (HF, 0.15 - 0.4 Hz) power through fast Fourier transform (FFT).

[0050] Import the time domain features and frequency domain features into the LSTM model to generate a state prediction score and a state prediction trend;

[0051] The LSTM model in the present invention is set as follows:

[0052] (1) LSTM network structure

[0053] Input layer: Receives the HRV feature sequence segmented by the sliding window, and the input dimension is (batch_size, window_size, num_features) (the number of features = 3, including SDNN, RMSSD, LF / HF).

[0054] Hidden layer: 2 layers of LSTM units, with 64 neurons in each layer, using the tanh activation function.

[0055] Output layer: A fully connected layer, outputting two branch results:

[0056] Regression branch: Predicts the inflammation score (continuous value).

[0057] Classification branch: Predicts the inflammation trend (deterioration / remission, binary classification).

[0058] (2) Loss function design

[0059] Combined loss function: Combines the mean squared error (MSE) of the regression task and the cross entropy (CE) of the classification task:

[0060]

[0061] Adjust the frequency and current intensity of the electrical stimulation according to the state prediction score, and adjust the pulse width according to the state prediction trend.

[0062] Adjust the frequency f and current intensity I parameters according to the following formula:

[0063] I new =I base +k l ·ΔS

[0064] f new = f base + k f ·ΔS

[0065] where ΔS is the scoring increment, and k l , k f is the gain coefficient.

[0066] If the predicted trend is "deterioration", the pulse width is preferentially increased: PW new = PW base × 1.2.

[0067] where I base , f base are the base current intensity and base frequency, which are based on the patient's weight, age, base heart rate, HRV baseline value (such as the initial value of SDNN), and personalized initial parameters are calculated through a multiple regression equation. The calculation method is as follows:

[0068] Base current intensity I base :

[0069] I base = α·weight + β·SDNN 初始 + γ

[0070] Base frequency f base :

[0071] f base = λ·age -1 + μ·RMSSD 初始

[0072] where the coefficients α, β, γ, λ, μ are all constants obtained through training with clinical data, and SDNN 初始 and RMSSD 初始 are the SDNN and RMSSD corresponding to the physiological state information of the patient collected for the first time.

[0073] The scoring increment ΔS = the state prediction score - the preset scoring threshold. For example, when the preset scoring threshold is 7.5, if the current score is 8.0, then ΔS = 0.5.

[0074] For the increment coefficient k l , k f , if the scoring change rate indicates that the patient's condition is deteriorating rapidly, the increment coefficient should be dynamically increased at this time, and the current intensity and frequency should be increased relatively high. For example, k l = k l + 0.05ΔS. If ΔS / Δt < -θ, it indicates that the patient's condition is improving rapidly, and the gain coefficient can be gradually reduced at this time to avoid overshoot.

[0075] Through a closed-loop mechanism of dynamic regulation, through the closed-loop process of real-time HRV monitoring → feature extraction → model inference → parameter adjustment → re-monitoring, real-time dynamic response is achieved: the electroacupuncture parameters are updated every 5 seconds to ensure dynamic matching with the inflammatory state, so that the electroacupuncture stimulation is in a state of dynamic adjustment throughout the treatment process and always provides appropriate electrostimulation to the patient, without causing side effects of overstimulation or understimulation.

[0076] On the other hand, as Figure 2 shown, an embodiment of the present invention also provides a closed-loop electroacupuncture stimulation regulation system based on LSTM. This system uses the above-mentioned closed-loop electroacupuncture stimulation regulation method based on LSTM, and includes:

[0077] A monitoring module, which is used to collect the physiological state data of the patient;

[0078] An inference module, which is used to extract the time-domain features and frequency-domain features of the physiological state data, obtain the corresponding state prediction score and state prediction trend according to the time-domain features and frequency-domain features, and then calculate the adjustment amplitude required for the frequency, current intensity, and pulse width according to the state prediction score and state prediction trend;

[0079] An electroacupuncture stimulation module, which is used to adjust the electroacupuncture stimulation parameters according to the adjusted frequency, current intensity, and pulse width.

[0080] The monitoring module is used to monitor heart rate variability (HRV), and real-time collects the patient's heart rate data through a single-lead electrocardiogram sensor with patch-type dry electrodes, and transmits the original ECG electrical signal to the data processing and inference module; in the present invention, the monitoring module includes: a multi-channel patch-type electrode element, a signal processing circuit; the multi-channel patch-type electrode element, the signal processing circuit, and the wireless module are integrated on the same main control board, and in the present invention, an electronic component AD8232 development kit can be used; in the present invention, the multi-channel electrodes are multiple high-precision and low-noise flexible electrode sheets, which are closely attached to the skin of the human abdominal infection site for collecting the electrocardiogram signal of the human body. The signal processing circuit includes a filtering circuit and an amplifying circuit. The filtering circuit is used to initially filter out high-frequency or low-frequency noise in the electrocardiogram signal, such as myoelectric interference, electromagnetic interference, and respiratory fluctuations. The amplifying circuit is used to amplify these weak electrocardiogram signals to a level suitable for processing, to avoid information loss caused by signal attenuation during transmission.

[0081] In the present invention, the electroacupuncture stimulation module adjusts the current intensity, frequency, and pulse width parameters of the electroacupuncture device according to the control signal generated by the inference module. The electroacupuncture stimulation module in the present invention includes a plurality of electroacupuncture electrodes, which are connected to the electroacupuncture stimulation module through wires and are given electrostimulation excitation by the electroacupuncture stimulation module. These electrostimulation signals include, but are not limited to, waveforms such as square waves, triangular waves, and sine waves, and parameters such as their frequency, intensity, and pulse width can be adjusted in real time.

[0082] On the other hand, as Figure 2 shown, an embodiment of the present invention further provides a computer, including at least one processor and a memory, where the memory stores a computer program, and the computer program is configured to be executed by the processor to implement the above-mentioned closed-loop electroacupuncture stimulation regulation method based on LSTM.

[0083] On the other hand, as Figure 2 shown, an embodiment of the present invention further provides a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. The computer program can be executed by one or more processors to implement the above-mentioned closed-loop electroacupuncture stimulation regulation method based on LSTM.

[0084] 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 the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A closed-loop electroacupuncture stimulation control method based on LSTM, characterized in that: include: Acquire the patient's physiological state data, and extract the time domain features and frequency domain features of the physiological state data; Import time domain features and frequency domain features into the LSTM model to generate state prediction scores and state prediction trends; The frequency and current intensity of electrical stimulation are adjusted according to the state prediction score, and the pulse width is adjusted according to the state prediction trend.

2. A closed-loop electroacupuncture stimulation control method based on LSTM as claimed in claim 1, characterized in that: After obtaining the physiological state data, the physiological state data needs to be preprocessed, and then the physiological state data is segmented using a sliding window to obtain multiple data sequences, and the time domain features and frequency domain features of each data sequence are extracted.

3. The closed-loop electroacupuncture stimulation control method based on LSTM as claimed in claim 1, characterized in that: The size of the sliding window is adjusted according to the prediction score obtained last time, and the sliding step size of the sliding window is adjusted according to the state prediction trend obtained last time.

4. The closed-loop electroacupuncture stimulation control method based on LSTM as claimed in claim 1, characterized in that: The time domain features include SDNN and RMSSD, and the frequency domain features include LF / HF. Then SDNN, RMSSD and LF / HF are combined into a three-dimensional input sequence and imported into the LSTM model.

5. The closed-loop electroacupuncture stimulation control method based on LSTM as claimed in claim 1, characterized in that: Methods for adjusting the frequency and current intensity of electrical stimulation based on the state prediction score include: Set the basic frequency and basic current intensity according to the patient's physiological state; The difference between the state prediction score and the preset score threshold is taken as the score increment, and then the score change rate of the score increment per unit time is calculated; A gain factor is determined according to the score change rate, and then the base frequency and base current intensity are increased according to the gain factor and the score increment.

6. A closed-loop electroacupuncture stimulation control method based on LSTM as claimed in claim 1, characterized in that: The setting formula of the basic current intensity is as follows: I base =a·weight+b·SDNN 初始 +g The setting formula of the basic frequency is as follows: f base =λ·age -1 +μ·RMSSD 初始 Among them, the coefficients α, β, γ, λ, and μ are all constants obtained by training with clinical data. The SDNN initial and RMSSD initial are the SDNN and RMSSD corresponding to the physiological state information of the patient collected for the first time.

7. A closed-loop electroacupuncture stimulation control system based on LSTM, characterized in that: The system uses a closed-loop electroacupuncture stimulation control method based on LSTM as described in any one of claims 1 to 6, comprising: A monitoring module, wherein the monitoring module is used to collect physiological status data of the patient; An inference module, which is used to extract time domain features and frequency domain features of physiological state data, and obtain corresponding state prediction scores and state prediction trends according to the time domain features and frequency domain features, and then calculate the amplitude of the frequency, current intensity and pulse width that needs to be adjusted according to the state prediction scores and state prediction trends; The electroacupuncture stimulation module is used to adjust the electroacupuncture stimulation parameters according to the adjusted frequency, current intensity and pulse width.

8. A computer, characterized in that: It includes at least one processor and a memory, wherein the memory stores a computer program, and the computer program is configured to be executed by the processor to implement a closed-loop electroacupuncture stimulation control method based on LSTM as described in any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, on which a computer program is stored. The computer program can be executed by one or more processors to implement a closed-loop electroacupuncture stimulation control method based on LSTM as described in any one of claims 1 to 6.

Images