A multi-modal fusion phase adaptive functional electrical stimulation rehabilitation system and method

Abstract

This method discloses a phase-adaptive FES modulation system based on multimodal signal fusion. It integrates brainwave motor imagery (BCI) signals, treadmill mechanical sensing signals (pedal angle, rotation speed, resistance), and electromyography (EMG) signals to construct a dynamic FES modulation model based on the lower limb pedaling motion phase (initiation / force exertion / retraction). Through AI algorithms, it decodes the intensity of motor intention and muscle activation state in real time, adaptively adjusting FES stimulation parameters (intensity, frequency, pulse width) to achieve precise closed-loop stimulation with a triple match of "intention-phase-muscle strength." This system ensures that stimulation is applied only during physiologically appropriate phases, avoiding motor interference and significantly improving the effectiveness and safety of lower limb rehabilitation training after stroke.

Description

Technical Field

[0001] This method belongs to the interdisciplinary field of intelligent rehabilitation engineering and neuroengineering. Specifically, it involves a phase-adaptive functional electrical stimulation (FES) modulation system based on multimodal signal fusion, which is particularly suitable for treadmill-assisted rehabilitation training scenarios for patients with lower limb motor dysfunction caused by stroke, spinal cord injury, etc. Background Technology

[0002] Functional electrical stimulation (FES) activates paralyzed or weakened lower limb muscles through the application of an external electrical current and has been widely used in neurorehabilitation therapy. However, existing FES systems still face the following key bottlenecks in clinical application:

[0003] The timing of stimulation is disconnected from the biomechanics of exercise: Traditional FES often uses fixed-sequence triggering (such as stimulating once per second) or relies solely on electromyographic threshold triggering, which cannot identify the specific biomechanical stage of the lower limb pedaling movement. For example, applying strong stimulation to the quadriceps during the treadmill recovery phase (knee flexion return phase) will antagonize the natural flexor muscle activity, not only reducing exercise efficiency but also easily causing muscle fatigue or even joint damage.

[0004] The stimulation parameters are static and lack individual dynamic adaptation: The stimulation intensity, frequency, pulse width, and other parameters of current devices are usually preset by the therapist and cannot be dynamically adjusted according to the patient's real-time central motor intention intensity (from the brain) and the actual activation state of peripheral muscles (from electromyography). When the patient's intention is strong but the nerve conduction is impaired, resulting in weak muscle strength, fixed parameters may not provide sufficient stimulation; conversely, when the muscles are already fully activated, overstimulation will result in energy waste and discomfort.

[0005] Intention perception is one-sided and the "central-peripheral" closed loop is not achieved: Some systems attempt to introduce brain-computer interfaces (BCIs) to decode motor imagery EEG signals to trigger FES, but such solutions generally have two major drawbacks: First, they do not integrate the mechanical state information of the treadmill, so they cannot determine whether the current motor phase is suitable for applying stimulation; second, they do not combine electromyographic feedback to verify the stimulation effect, resulting in the "intention → stimulation" process being an open loop, and it is impossible to confirm whether the stimulation effectively induces the target muscle contraction.

[0006] Lack of safety redundancy mechanisms: Existing systems rarely integrate real-time safety monitoring based on biomechanical fluidity or muscle fatigue status, making it difficult to cope with the accumulation of fatigue or fluctuations in active participation of patients during long-term training, posing potential safety risks.

[0007] Although existing research has attempted to use multi-source signals for FES control, no system has yet been able to simultaneously integrate three heterogeneous signals—EEG (intention), EMG (muscle strength), and mechanosensory (phase)—and construct a dynamic control model based on motion phase as the temporal reference and intention and muscle strength as the regulatory basis. Therefore, a novel FES control architecture is urgently needed to achieve truly precise stimulation through a triple match of "intention-phase-muscle strength". Summary of the Invention

[0008] To address the aforementioned issues, this method provides a phase-adaptive FES modulation system based on multimodal signal fusion. This system resolves problems such as inaccurate stimulation timing, rigid parameters, incomplete intent perception, and lack of closed-loop control in existing technologies. By constructing a dynamic modulation model based on the phase of lower limb pedaling motion, it achieves precise electrical stimulation that is highly coordinated with the patient's real-time physiological state, thereby improving the effectiveness, safety, and personalization of rehabilitation training.

[0009] To achieve the above objectives, the technical solution adopted in this method is as follows:

[0010] A phase-adaptive FES control system for multimodal signal fusion, the specific implementation of which includes the following steps:

[0011] Step 1: Synchronous acquisition of multimodal physiological and mechanical signals

[0012] The brainwave motor imagery (BCI) signals of the motor cortex (C3 / C4 / FCz region) are collected using a wearable EEG device; surface electromyography (EMG) signals of the rectus femoris, vastus lateralis, and medial head of the gastrocnemius muscle are collected using bipolar surface electrodes; and real-time pedal angle (0–360°), angular velocity (° / s), and resistance torque (N·m) signals are simultaneously acquired using a high-precision encoder and magnetic resistance unit integrated into the treadmill.

[0013] Step 2: Real-time phase division of lower limb pedaling motion

[0014] Based on the pedal angle θ, a single complete pedaling cycle is divided into three consecutive and mutually exclusive biomechanical phases:

[0015] The pedaling initiation phase (0° ≤ θ < 60°): corresponds to the hip and knee flexion preparation phase;

[0016] The force-generating phase (60° ≤ θ < 180°): corresponds to the main thrust phase of hip and knee extension;

[0017] Recovery phase (180° ≤ θ < 360°): corresponds to the hip and knee flexion return phase.

[0018] Step 3: Quantifying Exercise Intent and Muscle Activation State

[0019] Time-frequency analysis was performed on the BCI signal to extract event-related desynchronization (ERD) features of the mu rhythm (8–12 Hz) and beta rhythm (13–30 Hz). ERD reflects the activation level of the sensorimotor cortex during motor imagery, and its calculation formula is as follows:

[0020]

[0021] in, The average power spectral density of frequency band λ in the resting state The average power spectral density during the motion imagination task. This feature is input into a lightweight convolutional neural network (CNN), which decodes and outputs the motion intent intensity index I∈[0,1].

[0022] The EMG signal was subjected to full-wave rectification, 5 Hz low-pass filtering, and RMS calculation, and normalized to the individual's historical maximum activation level, outputting the muscle activation state index M∈[0,1]. The RMS value was calculated using the following formula:

[0023]

[0024] In the formula, Let N be the rectified electromyographic signal amplitude at the i-th sampling point, and N be the total number of sampling points within the sliding window (e.g., a 200 ms window corresponds to 200 points at 1000 Hz).

[0025] Step 4: Dynamically generate FES parameters based on triple matching

[0026] The current motion phase label P∈{start, exert, retraction}, intention index I, and muscle strength index M are input into a pre-trained multilayer perceptron (MLP) model, which outputs the FES stimulus parameter combination (S,F,W) in real time, where S is the stimulus intensity (mA), F is the frequency (Hz), and W is the pulse width (μs).

[0027] Step 5: Phase-adaptive electrical stimulation execution and safety feedback

[0028] Based on the parameters output from step 4, apply asymmetric biphasic square wave electrical stimulation to the target muscle group, and dynamically adjust according to the phase:

[0029] In the power generation phase, if I>0.6 and M<0.7, then increase S to 15–20 mA and F to 40–50 Hz;

[0030] In the starting phase, S is moderately increased to 8–12 mA to overcome static friction;

[0031] In the recovery phase, reduce S to 1–3 mA or turn off stimulation.

[0032] Simultaneously, the median EMG frequency decay rate and the coefficient of variation (CV) of treadmill angular velocity were continuously monitored to assess training safety. CV was calculated using the following formula:

[0033]

[0034] In the formula, and These represent the mean and standard deviation of the pedal angular velocity sequence over the most recent 2 seconds. If the median EMG frequency decreases by more than 15% or CV > 0.25 within 10 seconds, it is determined to be muscle fatigue or motion lag. The system automatically triggers stimulus decay or pause and uses the feedback data to optimize the online model.

[0035] Furthermore, the MLP model described in step 4 employs a transfer learning strategy: first, a general "phase-FES" mapping model is trained on data from a group of healthy subjects, and then a small number of samples are used for fine-tuning based on the training data from the first 3-5 minutes of patients to achieve rapid personalized adaptation.

[0036] Furthermore, in step 5, the negative phase pulse width of the asymmetric biphasic square wave is set to 1.2–1.5 times that of the positive phase, preferably 1.3 times, to reduce skin irritation and improve deep muscle recruitment efficiency.

[0037] Furthermore, all signal sampling rates do not exceed 1000 Hz, the total number of AI model parameters is less than 50k, and real-time inference can be achieved on embedded platforms (such as STM32H7 with TensorFlow Lite Micro), meeting the deployment needs of clinical or home rehabilitation scenarios.

[0038] Compared with existing technologies, the beneficial effects of this method are:

[0039] This method proposes a phase-adaptive FES modulation system based on multimodal signal fusion. By synchronously fusing three heterogeneous signals—EEG motor intention, EMG activation state, and treadmill mechanical phase—it constructs a dynamic modulation mechanism centered on "intention-phase-muscle strength." For the first time, it achieves synergistic optimization of FES stimulation across the time dimension (precise phase synchronization), intensity dimension (dual drive of intention and muscle strength), and safety dimension (closed-loop monitoring of fatigue and fluency). This system overcomes the limitations of traditional FES systems with fixed timing, single signal source, and open-loop control. It ensures that electrical stimulation is applied only during physiologically appropriate phases and adaptively adjusts parameters according to the patient's real-time state, significantly improving the accuracy, effectiveness, and safety of lower limb rehabilitation training. It also possesses good engineering feasibility and clinical application value. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the overall system architecture of this method;

[0041] Figure 2 A schematic diagram of the three phases of the lower limb pedaling cycle and the corresponding muscle group activation patterns;

[0042] Figure 3 Flowchart for multimodal signal fusion and FES dynamic control;

[0043] Figure 4 This is a schematic diagram illustrating the input-output mapping relationship of the AI ​​triple matching decision module.

[0044] Figure 5 The hardware circuit block diagram of the phase-adaptive FES execution module;

[0045] Figure 6 A schematic diagram of a typical rehabilitation training scenario for patients using this system.

Claims

1. A multimodal fusion-based phase-adaptive functional electrical stimulation rehabilitation method, characterized in that, Includes the following steps: Step 1: Synchronously acquire the patient's multimodal signals, including EEG motor imagery signals, surface electromyography signals, and treadmill mechanical sensing signals; Step 2: Based on the pedal angle in the mechanical sensor signal of the treadmill, the lower limb pedaling motion cycle is divided into three biomechanical stages in real time: the starting phase, the force exertion phase, and the recovery phase. Step 3: Decode the EEG motor imagery signal to obtain the motor intention intensity index, and process the surface electromyography signal to obtain the muscle activation state index; Step 4: Using the current motion phase, the motion intention intensity index, and the muscle activation state index as inputs, a functional electrical stimulation parameter combination is dynamically generated through an artificial intelligence model. The functional electrical stimulation parameter combination includes stimulation intensity, frequency, and pulse width. Step 5: Apply electrical stimulation to the target muscle group in the corresponding motor phase according to the functional electrical stimulation parameter combination, and monitor the training safety in real time. When a safety risk is detected, trigger the protection mechanism.

2. The multimodal fusion phase-adaptive functional electrical stimulation rehabilitation method as described in claim 1, characterized in that, In step 2, the division of the three biomechanical stages is as follows: Pedal start phase: The pedal angle θ satisfies 0° ≤ θ < 60°; Force application phase: The pedal angle θ satisfies 60° ≤ θ < 180°; Recovery phase: The pedal angle θ satisfies 180° ≤ θ < 360°.

3. The multimodal fusion phase-adaptive functional electrical stimulation rehabilitation method as described in claim 1, characterized in that, In step 3, the method for obtaining the motor intention intensity index is as follows: perform time-frequency analysis on the EEG motor imagery signal, extract the event-related desynchronization features of the mu rhythm (8–12 Hz) and beta rhythm (13–30 Hz), and input the features into a one-dimensional convolutional neural network for decoding; the method for obtaining the muscle activation state index is as follows: perform full-wave rectification on the surface electromyography signal, filter it using a second-order Butterworth low-pass filter with a cutoff frequency of 5 Hz, calculate the root mean square value within a 200 ms sliding window, and normalize it to the individual's historical maximum activation level.

4. The multimodal fusion phase-adaptive functional electrical stimulation rehabilitation method as described in claim 1, characterized in that, In step 4, the artificial intelligence model is a multilayer perceptron model, whose network structure includes an input layer, a hidden layer and an output layer, wherein the input layer has a dimension of 3, the hidden layer has a dimension of 16 and the output layer has a dimension of 3.

5. The multimodal fusion phase-adaptive functional electrical stimulation rehabilitation method as described in claim 1, characterized in that, In step 5, the specific strategy for applying electrical stimulation in the corresponding motion phase is as follows: During the exertion phase, if the intensity index of the movement intention is greater than 0.6 and the muscle activation index is less than 0.7, the stimulation intensity should be increased to 15–20 mA and the frequency to 40–50 Hz. During the initiation phase, the stimulation intensity is set to 8–12 mA; During the recovery phase, the stimulation intensity is reduced to 1–3 mA or stimulation is turned off.

6. The multimodal fusion phase-adaptive functional electrical stimulation rehabilitation method as described in claim 1, characterized in that, In step 5, the method for monitoring training safety is to continuously calculate the median frequency decay rate and the coefficient of variation of the treadmill angular velocity of the surface electromyography signal. If the median frequency drops by more than 15% or the coefficient of variation of the angular velocity is greater than 0.25 within 30 consecutive seconds, a safety risk is identified.

7. The multimodal fusion phase-adaptive functional electrical stimulation rehabilitation method as described in claim 4, characterized in that, The multilayer perceptron model is trained using a transfer learning strategy: first, a general model is trained on data from a group of healthy subjects, and then the output layer weights of the general model are fine-tuned based on the initial 5 minutes of training data from patients to achieve personalized adaptation.

8. A multimodal fusion phase-adaptive functional electrical stimulation rehabilitation system, characterized in that, include: The multimodal signal acquisition module is configured to simultaneously acquire the patient's brainwave motor imagery signals, surface electromyography signals, and treadmill mechanical sensing signals; The motion phase recognition module is configured to divide the lower limb pedaling motion cycle into a starting phase, a force exertion phase, and a recovery phase in real time based on the pedal angle in the mechanical sensor signal of the treadmill. The adaptive control module is configured to dynamically generate a combination of functional electrical stimulation parameters based on the motion phase, the intensity index of the motion intention, and the muscle activation state index. The phase-adaptive FES execution module is configured to apply electrical stimulation to the target muscle group in the corresponding motion phase according to the functional electrical stimulation parameter combination. The closed-loop safety feedback unit is configured to monitor training safety in real time and send protection commands to the adaptive control module and / or the phase adaptive FES execution module when a safety risk is detected.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the multimodal fusion phase-adaptive functional electrical stimulation rehabilitation method as described in any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multimodal fusion phase-adaptive functional electrical stimulation rehabilitation method as described in any one of claims 1 to 7.

Images