CN-122006102-A - Swallowing bionic electrical stimulation time sequence generation method based on multi-mode BCI and storage medium

CN122006102ACN 122006102 ACN122006102 ACN 122006102ACN-122006102-A

Abstract

The swallowing electric stimulation time sequence generation method based on the multi-mode BCI and the bionic time sequence electric stimulation is provided, a first swallowing electric stimulation sequence template is selected from a standard physiological swallowing muscle activation time sequence model based on individual patient information, a second swallowing electric stimulation sequence template is obtained through a reinforcement learning algorithm, the swallowing intention of a patient is identified through the multi-mode BCI, electric stimulation instructions are sent to a plurality of target muscle groups according to a preset time sequence according to the second swallowing electric stimulation sequence template, and parameters of the second swallowing electric stimulation sequence template are optimized and updated through the reinforcement learning algorithm to obtain a third swallowing electric stimulation sequence template. Through personalized treatment scheme, real-time feedback and self-adaptive adjustment, accurate swallowing intention recognition and continuously optimized treatment effect, the problems existing in the traditional swallowing rehabilitation treatment are solved, and the treatment effect and the rehabilitation speed of a patient are remarkably improved.

Inventors

ZHANG XINYU
WANG LIPING
LIU XIAOXUAN

Assignees

北京大学第三医院（北京大学第三临床医学院）

Dates

Publication Date: 20260512
Application Date: 20251127

Claims (10)

1. The swallowing bionic electrical stimulation time sequence generation method based on the multi-mode BCI is characterized by comprising the following steps of: Establishing a standard physiological swallowing muscle activation time sequence model based on the surface myoelectricity sEMG signals of the swallowing related muscles of the healthy person; selecting a first swallowing electrical stimulation sequence template from the standard physiological swallowing muscle activation timing model based on patient individual information; Acquiring physiological signals of the patient using the first swallowing electric stimulation sequence template in real time, and obtaining a second swallowing electric stimulation sequence template through a reinforcement learning algorithm; Identifying a patient swallowing intent via a multi-modal BCI, and transmitting electrical stimulation instructions to a plurality of target muscle groups in a predetermined time sequence according to the second swallowing electrical stimulation sequence template; and acquiring swallowing effect feedback data of the patient by using the second swallowing electric stimulation sequence template, and optimizing and updating parameters of the second swallowing electric stimulation sequence template by using a reinforcement learning algorithm to obtain a third swallowing electric stimulation sequence template.
2. The method of claim 1, wherein the establishing a standard physiological swallowing muscle activation timing model based on the surface myoelectricity sEMG signal of the swallowing-related muscle of the healthy person specifically comprises: collecting sEMG signals of mandibular hyoid muscle, anterior abdomen of two abdominal muscles, geniohyoid muscle, formative hyoid muscle and arytenoid muscle, and collecting laryngeal acceleration and swallowing sound signals; preprocessing the surface myoelectricity sEMG signal to extract a smooth sEMG envelope; Obtaining a muscle activation timing analysis and an inter-muscle timing relationship based on the smoothed sEMG envelope, laryngeal acceleration, and a deglutition tone signal; A standard physiological swallowing muscle activation timing model is established based on the muscle activation timing analysis and the inter-muscle timing relationship analysis.
3. The method of claim 2, wherein said obtaining a muscle activation timing analysis and an inter-muscle timing relationship based on said smoothed sEMG envelope, laryngeal acceleration and deglutition tone signal comprises: extracting activation start time, peak time and duration of mandibular hyoid muscle, two abdominal muscles anterior abdomen, geniohyoid muscle, hyoid muscle and arytenoid muscle based on the smooth sEMG envelope curve and the laryngeal acceleration signal y-axis peak point, wherein the activation start time, peak time and duration form the muscle activation time sequence analysis; Normalizing the smoothed sEMG envelopes of any two groups of deglutition-related muscles, and calculating a cross-correlation function of the sEMG envelopes of the two groups of deglutition-related muscles Obtaining the time sequence relation among the muscles, wherein For the normalized muscle sEMG envelope, T is the swallowing duration, Is time delay.
4. The method of claim 1, wherein selecting a first swallowing electrical stimulation sequence template from the standard physiological swallowing muscle activation timing model based on patient individual information specifically comprises: Acquiring physiological information of the patient, wherein the physiological information at least comprises one or more of laryngeal lift, swallowing tone, and Upper Esophageal Sphincter (UES) opening; A first swallowing electrical stimulation sequence template is selected from the standard physiological swallowing muscle activation timing model using a weighted euclidean distance model based on the patient physiological information.
5. The method of claim 4, wherein the acquiring in real time physiological signals of the patient using the first swallowing electrical stimulation sequence template, the obtaining a second swallowing electrical stimulation sequence template by a reinforcement learning algorithm specifically comprises: Acquiring a first physiological signal of the patient using the first swallowing electrical stimulation sequence template, the first physiological signal comprising an sEMG signal of the swallowing-related muscle, as well as laryngeal acceleration and a swallowing sound signal; preprocessing the first physiological signal, and extracting key characteristics of the first physiological signal; optimizing the first swallowing electric stimulation sequence template by adopting a depth deterministic strategy gradient, taking key features of a first physiological signal as input signals, and outputting the key features as first electric stimulation parameters after optimizing the first swallowing electric stimulation sequence template; the second swallowing electrical stimulation sequence template is obtained based on the first electrical stimulation parameter.
6. The method of claim 5, wherein optimizing the first swallowing electrical stimulation sequence template using depth deterministic strategy gradients, outputting first physiological signal key features as input signals as first electrical stimulation parameters optimized for the first swallowing electrical stimulation sequence template specifically comprises: loading the first swallowing electrical stimulation sequence template parameters as current parameters , Wherein As a function of the current stimulation time difference, Is of initial value For the current stimulus intensity, the magnitude of the current applied to the muscle by the electrical stimulus, For the initial value of the value, For the current stimulation frequency, Is an initial value; building State vectors The State vector S t contains the target muscle activation strength Ratio of throat lifting amplitude to standard value Choking cough marker C, food cluster residual score And current electrical stimulation parameters , ; Performing an Actor network cycle training to obtain at least 256 sample numbers, and calculating Critic network loss based on the at least 256 sample numbers; adopting an Adam optimizer to perform back propagation optimization to respectively update the Critic network and the Actor network; And extracting 3 groups of parameters with highest rewards, and calculating an average value to be used as an updating parameter of the first swallowing electric stimulation sequence template, so as to obtain the second swallowing electric stimulation sequence template.
7. The method of claim 1, wherein the identifying the patient's swallowing intent via the multimodal BCI and the sending the electrical stimulation instructions to the plurality of target muscle groups in a predetermined time sequence according to the second swallowing electrical stimulation sequence template specifically comprises: the patient sends out a swallowing command through a cortical brain electrical signal adopted by motor imagery or steady-state visual evoked potential; the BCI system identifies the swallowing instructions, invokes the second swallowing electrical stimulation sequence template for the patient; And sending instructions to the electric stimulation units of a plurality of channels according to the second swallowing electric stimulation sequence template, and sequentially stimulating the swallowing related muscles.
8. The method of claim 1, wherein the acquiring the patient swallowing effect feedback data using the second swallowing electrical stimulation sequence template, optimizing updating parameters of the second swallowing electrical stimulation sequence template using a reinforcement learning algorithm, and obtaining a third swallowing electrical stimulation sequence template specifically comprises: acquiring swallowing data of the patient using the second swallowing electrical stimulation sequence template to obtain a second physiological signal; The second physiological signal is preprocessed, the standard chemometric feature vector is extracted, and the difference value between the target standardized feature vector value and the actual standardized feature vector value is calculated to obtain a difference value standardized feature vector And based on the difference normalized feature vector, performing closed-loop feedback and adaptive learning on the second swallowing electric stimulation sequence template to obtain the third swallowing electric stimulation sequence template.
9. The method of claim 8, wherein the performing closed loop feedback and adaptive learning on the second swallowing electrical stimulation sequence template based on the difference normalized feature vector to obtain the third swallowing electrical stimulation sequence template specifically comprises: Calculating pearson correlation coefficients of stimulation parameters and curative effect characteristics of swallowing-related muscles Screening stimulus parameters with absolute values of correlation coefficients larger than a threshold value as sensitive parameters; taking the difference normalized feature vector and the third swallowing electrical stimulation sequence template as a current state S; selecting an action to maximize Q (S, A) Wherein Q (S, a) represents an action cost function of taking action a in state S; action of maximizing Q (S, A) As parameters for updating the second swallowing electrical stimulation sequence template.
10. A computer-readable storage medium storing computer instructions for causing a computer to perform the swallowing biomimetic electrical stimulation timing generation method based on a multi-modal BCI as claimed in any one of claims 1-9.

Description

Swallowing bionic electrical stimulation time sequence generation method based on multi-mode BCI and storage medium Technical Field The invention relates to the field of brain-computer interface (BCI) and intelligent medical rehabilitation, in particular to a swallowing bionic electrical stimulation time sequence generation method based on multi-mode BCI. Background Many diseases can cause dysphagia in patients, particularly MND patients, which can potentially lead to fatal complications. Traditional swallowing rehabilitation has limited effect, particularly neuromuscular electrical stimulation (NMES) adopts fixed parameters and modes to stimulate peripheral muscles, and has the core defect that the peripheral muscles are disjointed with central intention, the stimulation is triggered by equipment at fixed time, rather than the swallowing intention of a patient, and closed-loop nerve remodeling cannot be formed. The non-physiological stimulation mode usually adopts synchronous stimulation, which is inconsistent with the physiological rule of sequential and coordinated contraction of normal swallowing muscles, and has limited effect and possibly aggravates the incompatibility. The parameters are solidified, and self-adaptive adjustment can not be carried out according to individual differences of patients and disease progress. Although the existing BCI technology attempts to trigger stimulation with "ideas", most of them only implement a simple "switch" function, and do not solve the deeper problem of how to stimulate more accurately and more in accordance with physiological laws. The present invention is based on the insight that the physiological and pathological mechanisms of swallowing are well understood, and that the use of AI technology translates this knowledge into an executable, personalized, accurate stimulation protocol. Although existing BCI techniques attempt to trigger stimulation with "ideas", most only implement a simple "switch" function, its application paradigm is still limited to the "trigger-recovery" mode, i.e., assuming that neural pathways can be reconstructed by repeated intent-stimulus pairing training. However, for patients with progressive, irreversible neurological diseases such as Motor Neuron Diseases (MND), the swallowing function cannot be truly recovered, and the effect of the conventional rehabilitation training mode is limited. These patients are in urgent need for a device that can "replace" or "augment" their impaired nerve function, and provide safe and effective swallowing assistance immediately when needed, without requiring permanent recovery of nerve function. The invention aims to develop a bionic 'neural prosthesis' or 'functional auxiliary appliance' based on the clinical requirement, and aims to convert the knowledge of a clinician on a swallowing physiological mechanism into an executable and personalized accurate stimulation scheme, and provide a section of perfect bionic muscle contraction sequence for a patient when the patient initiates intention, so as to compensate the lost function of the patient. Disclosure of Invention The invention aims to solve the problem that the stimulation mode and the physiological time sequence are disjointed and lack of individuation in the prior art, and provides a swallowing bionic electric stimulation time sequence generation method based on a multi-mode BCI, which is characterized by comprising the following steps of: Establishing a standard physiological swallowing muscle activation time sequence model based on the surface myoelectricity sEMG signals of the swallowing related muscles of the healthy person; selecting a first swallowing electrical stimulation sequence template from the standard physiological swallowing muscle activation timing model based on patient individual information; Acquiring physiological signals of the patient using the first swallowing electric stimulation sequence template in real time, and obtaining a second swallowing electric stimulation sequence template through a reinforcement learning algorithm; Identifying a patient swallowing intent via a multi-modal BCI, and transmitting electrical stimulation instructions to a plurality of target muscle groups in a predetermined time sequence according to the second swallowing electrical stimulation sequence template; and acquiring swallowing effect feedback data of the patient by using the second swallowing electric stimulation sequence template, and optimizing and updating parameters of the second swallowing electric stimulation sequence template by using a reinforcement learning algorithm to obtain a third swallowing electric stimulation sequence template. In a second aspect, the present invention provides a computer readable storage medium, where the computer readable storage medium stores computer instructions for causing a computer to execute the above-described method for generating swallowing bionic electrical stimulation based on a multi-modal