Electrical stimulation awakening treatment system and equipment based on unsupervised learning model

Abstract

The present disclosure relates to an electrical stimulation wakefulness treatment system and equipment based on an unsupervised learning model, which is used to solve the problem that the existing technology cannot timely and effectively promote personalized wakefulness for patients with narcolepsy. The system includes a first stimulation module and a second stimulation module; the first stimulation module is configured to continuously stimulate the patient's vagus nerve during the day; the second stimulation module is configured during the operation of the first stimulation module, based on the time-frequency characteristics of the EEG signal collected at the frontal central Fz position, using an unsupervised learning method to detect and classify the patient's state; when it is detected that the patient is in a sleeping or drowsy state, the patient's trigeminal nerve is pulsed stimulated until the patient is detected to be in a wakeful state. The above scheme is safe through non-invasive electrical stimulation wakefulness treatment, and can timely intervene in sudden daytime sleep, and the unsupervised learning method used can provide personalized wakefulness treatment for different patients without external labels.

Description

Technical Field

[0001] The present disclosure relates to the field of medical technology, and in particular to an electrical stimulation awakening treatment system based on an unsupervised learning model. Background Art

[0002] Narcolepsy is a chronic neurological disease whose main symptoms include uncontrollable daytime sleepiness and sudden falling asleep, which will increase the risk of emergencies and have a serious impact on the patient's work and life. At present, common treatment methods are mostly focused on drug therapy, but drug dependence and side effects have led to an increasing demand for non-drug treatments.

[0003] Existing technologies mainly focus on simply identifying wakefulness and sleep states through EEG signal monitoring, but do not use this to regulate or regulate electrical stimulators based on other parameters for real-time wakefulness-promoting treatment. For patients with narcolepsy, the timeliness and effectiveness are slightly insufficient. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present disclosure proposes an electrical stimulation wake-up treatment system based on an unsupervised learning model, which aims to combine the automatic classification of real-time EEG signals with electrical stimulation wake-up intervention to provide a new wake-up solution for narcolepsy patients. The specific technical solution is as follows.

[0005] On the first aspect, this case proposes an electrical stimulation awakening treatment system based on an unsupervised learning model, the system comprising a first stimulation module and a second stimulation module; the first stimulation module is configured to continuously stimulate the patient's vagus nerve during the day; the second stimulation module is configured during the operation of the first stimulation module, based on the time-frequency characteristics of the EEG signal collected at the frontal central Fz position, to detect and classify the patient's state using an unsupervised learning method; when the patient is detected to be in a sleeping or drowsy state, the patient's trigeminal nerve is electrically stimulated by pulses until the patient is detected to be awake.

[0006] In one implementation of the above technical solution, the EEG signal collected based on the fronto-central Fz position is preprocessed, and the preprocessed EEG data is subjected to a short-time Fourier transform;

[0007] The preprocessing is to use a Butterworth filter to perform 0.5 Hz high-pass filtering and 35 Hz low-pass filtering to remove low-frequency drift, electrocardiogram and electromyography artifacts to obtain EEG data.

[0008] In one implementation of the above technical solution, detection and classification are achieved through an unsupervised learning model. The unsupervised learning model first extracts local features based on the EEG signal collected at the fronto-central Fz position, and reduces the spatial size to achieve preliminary dimensionality reduction. After flattening the features of the preliminary dimensionality reduction, sequence features are obtained. The sequence features are then compressed into a vector of a fixed length in the time dimension to achieve time dimensionality reduction, and the patient state features are obtained. Finally, the patient state features are detected and classified.

[0009] In one implementation of the above technical solution, the unsupervised learning model is composed of an autoencoder model and a clustering algorithm.

[0010] In one implementation of the above technical solution, the autoencoder model is composed of a convolutional neural network and a long short-term memory network.

[0011] In one implementation of the above technical solution, the clustering algorithm is preferably a K-Means algorithm.

[0012] In one embodiment of the above technical solution, the first stimulation module implements stimulation on the inner and outer sides of the middle of the tragus at the entrance of the unilateral external auditory canal, and the stimulation parameters are current intensity 0.5mA, frequency 25Hz, stimulation time 30s, interval time 30s, and pulse width 200-300μs.

[0013] In one embodiment of the above technical solution, the second stimulation module is implemented in the supraorbital nerve, and automatically triggers a stimulation pulse when it is detected that the patient is in a sleep or drowsy state. The stimulation parameters are a current intensity in the range of 10-15 mA, a stimulation frequency of 40 Hz, and a pulse width of 200 μs.

[0014] In one implementation of the above technical solution, the EEG signal of the fronto-central FZ is collected using the left or right ear electrode as a reference, with a sampling rate of 512 Hz.

[0015] In a second aspect, the present disclosure proposes an electrical stimulation awakening therapy device based on an unsupervised learning model, characterized in that it includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and executed by any of the above systems.

[0016] Beneficial technical effects:

[0017] (1) Through EEG signal regulation, non-invasive electrical stimulation to promote wakefulness is implemented. Compared with existing narcolepsy treatment methods that mainly rely on drugs and are prone to drug side effects, this program is safe. However, it is difficult to provide timely intervention for sudden daytime sleepiness. Existing EEG monitoring methods are mostly used for analysis of sleep structure and lack direct integration with treatment methods. This program effectively utilizes the advantages of EEG technology in judging sleep status with high sensitivity and accuracy, thereby improving the effectiveness and timeliness of treatment.

[0018] (2) This scheme uses a machine learning model to construct an autoencoder to extract deep features of EEG signals. Compared with the existing threshold control method based on HRV heart rate variability parameters, this scheme can effectively capture subtle changes in the sleep state transition process, while automatically adapting to subtle differences in individual sleep patterns. The classification accuracy is improved, and the accuracy and specificity of the regulation stimulation therapy are also stronger.

[0019] (3) This scheme adopts an unsupervised learning model, which can effectively deal with the data differences between different individual patients without external labels. Even if the sleep pattern of the same individual may change over time, it can adapt to the data changes through dynamic clustering methods, thereby significantly improving the specificity and convenience of the overall treatment plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0021] Figure 1 , one Schematic diagram of the overall process in an implementation manner.

[0022] In the figure, 110-trigeminal nerve stimulation, 111-frontal area EEG signal collection, 112-time-frequency feature extraction and construction of autoencoder model dimensionality reduction, 113-unsupervised real-time clustering analysis, 114-judgment of wakefulness, 115-stop trigeminal nerve electrical stimulation, 116-trigger trigeminal nerve electrical stimulation, 120-vagus nerve stimulation, 130-wake-up treatment for patients with narcolepsy. DETAILED DESCRIPTION

[0023] For patients with narcolepsy, the symptoms of daytime sleepiness have a great impact on these patients and are difficult to control and respond to in a timely manner. They are also sudden and bring risks to patients' lives. This program proposes an electrical stimulation wake-up treatment system based on an unsupervised learning model to effectively help patients control the risks brought by the disease.

[0024] The following is a clear and complete description of how to implement the technical solution of this case. Obviously, the described implementation method is only a part of the implementation method of this case, not all of the implementation methods. Based on the implementation method in this case, all other implementation methods obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0025] 1. Data acquisition module

[0026] The data acquisition module is configured to collect single-channel EEG (Electroencephalogram) data from the frontal region of patients with narcolepsy. During the acquisition, the position of the frontal central Fz is determined according to the international 10-20 system, the reference electrode is the left or right ear electrode, and the sampling rate is 512Hz. The data is used to monitor the brain activity characteristics of patients in different states in real time.

[0027] 2. Signal preprocessing module

[0028] The collected frontal EEG signals were preprocessed by using a Butterworth filter for 0.5 Hz high-pass filtering and 35 Hz low-pass filtering to remove low-frequency drift, ECG and EMG artifacts, and obtain cleaner EEG signals for subsequent feature extraction.

[0029] (III) Signal processing module

[0030] The preprocessed EEG data is analyzed by short-time Fourier transform (STFT) to extract the time-frequency characteristics of the signal. The spectrum amplitude calculated by STFT provides energy distribution information in the entire frequency range.

[0031] Use the trained autoencoder model to reduce the dimension, so as to compress the high-dimensional input features into a low-dimensional representation for subsequent analysis. Perform real-time unsupervised cluster analysis on the reduced-dimensional feature data to classify the EEG signals into two states: awake or asleep. Compared with traditional supervised learning methods, this unsupervised learning method can avoid the intervention of clinicians in pre-manual classification when constructing a patient's private data set. While ensuring specificity and accuracy, it improves the efficiency of the overall process and ensures that users can operate and use it independently.

[0032] In one embodiment, the autoencoder model includes a CNN model and an LSTM model connected in series. First, the input time-frequency features are processed by a convolutional neural network (CNN). CNN extracts local features and reduces the spatial size through convolution and pooling operations, thereby achieving preliminary dimensionality reduction. Then, these features are flattened and input into a long short-term memory network (LSTM). LSTM further encodes the sequence features in the time dimension and compresses them into a vector of fixed length to achieve time dimensionality reduction. The entire process uses the spatial dimensionality reduction of CNN and the temporal dimensionality reduction of LSTM to ultimately obtain a low-dimensional representation containing the original information, while removing a large amount of redundancy and noise, providing a more effective feature representation for subsequent clustering or classification.

[0033] In one implementation, the cluster analysis uses a K-Means algorithm.

[0034] 4. Stimulus Control Feedback Module

[0035] The stimulation control feedback module includes two parts: the first stimulation module and the second stimulation module, which correspond to vagus nerve stimulation and trigeminal nerve stimulation respectively, and both use transcutaneous electrical stimulation. Among them: for vagus nerve stimulation, the positive and negative electrodes are placed on the inner and outer sides of the middle of the tragus at the entrance of the left (or right) external auditory canal (vagus nerve afferent fiber branch) for continuous stimulation during the day, and the stimulation parameters are set to current intensity of 0.5mA, frequency of 25Hz, stimulation time of 30s, interval time of 30s, and pulse width of 200-300μs. Trigeminal nerve stimulation acts on the supraorbital nerve, which passes through the supraorbital notch and branches are distributed on the forehead and top skin. Based on the real-time clustering results, this stimulation module determines whether the patient is in a state of sleep or sleepiness. If a state of sleep or sleepiness is detected, it automatically triggers a wake-up electrical stimulation pulse to stimulate the supraorbital nerve position until the awake EEG state is detected. Under safe conditions, the current intensity is in the range of 10-15 mA, the stimulation frequency is 40 Hz, and the pulse width is 200μs.

[0036] The above-mentioned daytime is the patient's activity time other than normal sleeping time. Normal sleeping time is the patient's sleep and rest time the same as normal people, not the patient's sudden sleep time.

[0037] The terms "first" and "second" mentioned above are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.

[0038] Figure 1Schematic diagram of the overall process in one embodiment. The stimulation performed during the wake-promoting treatment (130) for patients with narcolepsy includes two parts, one part is trigeminal nerve stimulation (110), and the other part is vagus nerve stimulation (120). Among them, the patient's vagus nerve is continuously stimulated during the day. When implementing trigeminal nerve stimulation, frontal area EEG signal acquisition (111), video feature extraction and construction of autoencoder model dimensionality reduction (112), and unsupervised real-time clustering analysis (113) are performed in sequence. According to the analysis results, it is determined whether the patient is awake (114). If the patient is in an awake state, the trigeminal nerve electrical stimulation is stopped (115), otherwise, the trigeminal nerve electrical stimulation is triggered (116).

[0039] In summary, this solution combines EEG monitoring and non-invasive electrical stimulation therapy to provide a more immediate and effective narcolepsy treatment system. Compared with other methods such as HRV fatigue index, this solution has a timely warning method and has stronger specificity and accuracy. It does not rely on manual labeling by experts and is more efficient. The system can monitor the patient's status in real time, intervene in time when daytime sleep attacks occur, and enhance the patient's daytime excitability by continuously stimulating the vagus nerve. The dual mode ensures the patient's quality of life and avoids drug side effects and long-term dependence. In addition, this solution uses non-invasive treatment methods, which is safer and suitable for long-term treatment and daily management.

[0040] Through the description of the above implementation methods, those skilled in the art can clearly understand that the disclosed system can be implemented by means of software plus necessary general hardware, and of course can also be implemented by dedicated hardware including dedicated integrated circuits, dedicated CPUs, dedicated memories, dedicated components, etc. In general, all functions performed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structures used to implement the same function can also be diverse, such as analog circuits, digital circuits, or dedicated circuits, etc. However, for the present disclosure, software program implementation is a better implementation method in most cases.

[0041] Although the embodiments of the present disclosure are described above in conjunction with the accompanying drawings, the present disclosure is not limited to the above-mentioned specific embodiments and application fields. For example, it can also be applied to anti-drowsiness, anti-driving fatigue, etc. The above-mentioned specific embodiments are only illustrative and instructive, but not restrictive. Under the guidance of this specification and without departing from the scope of protection of the claims of the present disclosure, ordinary technicians in this field can also make many forms, which all belong to the protection of the present disclosure.

Claims

1. An electrical stimulation awakening therapy system based on an unsupervised learning model, characterized in that: The system includes a first stimulation module and a second stimulation module; The first stimulation module is configured to continuously stimulate the vagus nerve between the inner and outer sides of the middle of the tragus at the entrance of the unilateral external auditory canal of the patient during the day, with stimulation parameters of current intensity of 0.5 mA, frequency of 25 Hz, stimulation time of 30 s, rest time of 30 s, and pulse width of 200-300 μs; The second stimulation module is configured during the operation of the first stimulation module to detect and classify the patient's state based on the time-frequency characteristics of the EEG signal collected at the frontal central Fz position using an unsupervised learning method; when the patient is detected to be in a sleeping or drowsy state, the patient's supraorbital trigeminal nerve is pulsed stimulated, and the stimulation parameters are a current intensity in the range of 10-15 mA, a stimulation frequency of 40 Hz, and a pulse width of 200 μs, until the patient is detected to be in an awake state.

2. The system according to claim 1, characterized in that: The EEG signals collected based on the fronto-central Fz position were preprocessed, and the preprocessed EEG data were subjected to short-time Fourier transform; The preprocessing is to use a Butterworth filter to perform 0.5 Hz high-pass filtering and 35 Hz low-pass filtering to remove low-frequency drift, electrocardiogram and electromyography artifacts to obtain EEG data.

3. The system according to claim 1, characterized in that: Detection and classification are achieved through an unsupervised learning model. The unsupervised learning model first extracts local features based on the EEG signal collected at the frontal central Fz position, and reduces the spatial size to achieve preliminary dimensionality reduction. After flattening the features of the preliminary dimensionality reduction, the sequence features are obtained. The sequence features are then compressed into a vector of a fixed length in the time dimension to achieve time dimensionality reduction, and the patient state features are obtained. Finally, the patient state features are detected and classified.

4. The system according to claim 3, characterized in that The unsupervised learning model is composed of an autoencoder model and a clustering algorithm.

5. The system according to claim 4, characterized in that The autoencoder model is composed of a convolutional neural network and a long short-term memory network.

6. The system according to claim 4, characterized in that The clustering algorithm is preferably a K-Means algorithm.

7. The system according to claim 1, characterized in that Features: The EEG signals of the frontocentral FZ were collected using the left or right ear electrodes as reference, with a sampling rate of 512 Hz.

8. An electrical stimulation awakening treatment device based on an unsupervised learning model, characterized in that: The system comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the system as claimed in any one of claims 1 to 7.

Images