Intelligent electrical stimulation site dynamic selection method based on electrode array

Abstract

The invention discloses an intelligent electrical stimulation site dynamic selection method based on an electrode array, and relates to the technical field of electrical stimulation medical rehabilitation training.The method comprises the steps that motion angle data, stimulation site information data and electrical stimulation parameter data of a wrist joint of a historical user are collected; training the random forest model according to a training data set established by wrist joint motion angle data and stimulation site information data of any specific action of the user; inputting user wrist joint motion angle data collected in real time into the trained random forest model to obtain a next effective stimulation site which needs to output electrical stimulation pulses; generating corresponding electrical stimulation pulses according to the effective stimulation sites, transmitting the electrical stimulation pulses to the array electrode sites, and inducing muscle contraction of the user to generate limb actions; according to the method, by changing wrist motion angle data input into a model, array electrode stimulation sites outputting electrical stimulation are adjusted in real time, so that the effect that wrist actions induced by electrical stimulation are kept stable and unchanged when an arm rotates is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical stimulation medical rehabilitation training, and in particular to an intelligent electrical stimulation site dynamic selection method based on an electrode array. Background Art

[0002] Loss of hand function in stroke survivors is primarily manifested in motor impairments. These impairments make it difficult for patients to perform basic movements such as grasping, lifting, or extending their fingers, severely reducing their ability to carry out activities of daily living. Timely rehabilitation training of arm muscles using rehabilitation equipment can help them regain basic hand function.

[0003] Functional electrical stimulation (FES) is a technique that promotes hand function rehabilitation by delivering electrical stimulation pulses via surface electrodes to activate neuromuscular tissue. The size, position, and placement of the electrodes significantly influence the selective muscle activation and the intensity of the induced limb movements. Relative movement between skeletal muscle and skin during changes in body position can alter the effectiveness of electrical stimulation, resulting in variations in the type or intensity of limb movements under the same electrical stimulation. To address the above problems, FES systems based on array electrodes can solve the need for constant re-application of electrodes. Several existing methods have been proposed: Malesevic et al. demonstrated a reliable calibration method by using fewer sensors (n=3), proving that calibration accuracy can be maintained while reducing equipment complexity; de Marchis C et al. and Knibbe et al. introduced electrophysiological drive methods for electrode calibration, combining the evaluation of joint angles and muscle contractions through electromyography, providing a more comprehensive basis for judging expected results; Salchow-Hömmen C et al. evaluated the application of semi-automatic and fully automatic calibration algorithms in the hand opening / closing task of stroke patients. Both methods were highly accepted by users, and patients did not show a clear preference for one method over the other. The semi-automatic method was 25% faster than the fully automatic method due to expert supervision and the use of human cycle optimization. This suggests that while pursuing calibration efficiency, it is also important to consider how to incorporate human expertise and experience into algorithm design. Salchow C et al. introduced a feedback control-assisted method that allows adjustment of the position of virtual electrodes in the array. Through interpolation functions, this method estimates neighboring elements and their stimulation parameters, thereby improving user integration and acceptance of the calibration algorithm. Guangyu Zhao et al. designed a multi-site electrode hand functional rehabilitation system that achieved precise position control of multi-joint hand movements by applying an iterative learning control (ILC) algorithm with a forgetting factor. Malešević et al. trained an artificial neural network (ANN) to classify different hand movements based on accelerometer data and compared them with standard goniometric data, demonstrating the feasibility of using a minimum number of sensors for calibration tasks. Imatz-Ojanguren et al. proposed a recursive fuzzy neural network (RFNN) for FES-controlled hand movements. The model takes electrode configuration and stimulation parameters as input and maps them to wrist and finger positions. Leveraging the linguistic interpretability of fuzzy logic and the feedback loop of the ANN, the RFNN model can predict the output response given the parameter inputs.

[0004] However, although the above-mentioned existing technologies use different methods to determine the optimal stimulation sites for producing different gesture movements, they are unable to dynamically select the determined stimulation sites to adapt to the ever-changing forearm posture and the needs of daily life activities; when the skin, skeleton and muscles move relative to each other, when the existing methods are used for functional electrical stimulation, the changing muscle response will lead to unstable and incoherent functional movements, thereby affecting the user's rehabilitation training effect. Summary of the Invention

[0005] In response to the deficiency in the existing technology that it is impossible to dynamically select and determine the stimulation sites, the present invention proposes a dynamic selection method for intelligent electrical stimulation sites based on an electrode array. By deploying a random forest model to analyze wrist motion angle data in real time to determine new stimulation sites, the problems existing in the existing technology are solved.

[0006] A method for dynamic selection of intelligent electrical stimulation sites based on an electrode array, comprising the following steps: Collecting wrist joint motion angle data and corresponding stimulation site information when the user makes limb movements; after applying an electrode array to the back and ventral sides of the user's arm, applying current pulses with fixed electrical stimulation parameters to each electrode site in the electrode array in sequence to determine whether the current electrode site can induce muscle contraction to produce limb movement; A training data set is established based on the movement angle data of the wrist joint and the stimulation site information data under any specific limb movement of the user, and the random forest model is trained using the training data set; The wrist joint motion angle data of the current user to be tested is input into the trained random forest model to obtain the next effective stimulation site to output the electrical stimulation pulse; based on the effective stimulation site, the corresponding electrical stimulation pulse is generated and transmitted to the electrode site of the electrode array, thereby inducing the user's muscle contraction to produce limb movement.

[0007] Furthermore, a functional electrical stimulation system with an electrode array configured as a monopolar electrode is used to apply electrical stimulation to the stimulation site to obtain the motion angle data of the user's wrist joint; the functional electrical stimulation system with an electrode array configured as a monopolar electrode specifically includes a stimulation electrode, a reference electrode, a biphasic current pulse sequence capable of generating constant parameters, and a Leap Motion somatosensory interactor; wherein, the stimulation electrode is each electrode site in the electrode array, the reference electrode is an electrode placed at the joint and having an area larger than the electrode site, and the Leap Motion somatosensory interactor is used to capture the motion angle data of the wrist joint under changes in arm posture after applying an electrode array to the back and ventral side of the user's arm.

[0008] Furthermore, the electrode array surface functional electrical stimulation system with a monopolar electrode configuration includes two 2×5 array electrodes and 20 stimulation channels; each electrode in the array electrode has a size of 20 mm×20 mm.

[0009] Furthermore, the Leap Motion somatosensory interactor uses binocular cameras and binocular stereo vision principles to build a three-dimensional hand model, and tracks and captures wrist joint motion angle data through the three-dimensional hand model.

[0010] Furthermore, the digital-to-analog converter DAC and the stimulation pulse output driving circuit are driven by the timer interrupt program to generate corresponding electrical stimulation pulses which are transmitted to the array electrode sites.

[0011] Furthermore, the random forest model is trained using the training data set, specifically comprising the following steps: The training data set is randomly shuffled and divided into ten equal parts by using the ten-fold cross-validation method. Each part of the data is used as the test set in turn, and the rest are used as the training set for random forest model prediction. Tune key parameters based on the average prediction accuracy of the random forest model; The training data set is input into the random forest model after parameter tuning, and multiple decision trees are generated through Bootstrap sampling. Each tree randomly selects a feature subset for node splitting to obtain the trained random forest model.

[0012] The present invention provides a method for dynamic selection of intelligent electrical stimulation sites based on an electrode array, which has the following beneficial effects: The present invention trains a random forest model by utilizing the motion angle data of the wrist joint when the user generates limb movements and the corresponding stimulation site information data, so as to determine a new stimulation site on the electrode array of the user's controlled target limb, and transmit the electrical stimulation pulse to the newly determined stimulation site, thereby ensuring that the functional movement task performed by the controlled forearm in the previous forearm posture will not be interrupted due to the change of the forearm posture; this method can dynamically adjust the array electrode stimulation site of the output electrical stimulation in real time by changing the wrist motion angle data input into the random forest model, so as to achieve the effect of keeping the wrist movement induced by electrical stimulation stable when the arm rotates. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a working principle diagram of the array electrode surface FES system in an embodiment of the present invention; Figure 2 This is a flowchart of the embedded program operation in an embodiment of the present invention; Figure 3This is a schematic diagram of the process of automatically finding the body surface FES point in the embodiment of the present invention and executing the "site recognition" instruction to automatically obtain the stimulation site in different body positions; Figure 4 Schematic diagram of the human-computer interaction feedback PC host system interface in an embodiment of the present invention; Figure 5 This is a flowchart of the operation of the PC human-computer interaction feedback system in an embodiment of the present invention; Figure 6 This is a flowchart of Leap Motion acquiring wrist motion angle data in an embodiment of the present invention; Figure 7 Schematic diagram of forearm muscle distribution and array electrode placement in an embodiment of the present invention; Figure 8 Schematic diagram of parameter value learning curve in an embodiment of the present invention; Figure 9 Graphs showing wrist extension / flexion motions of a controller and a controlled person in an embodiment of the present invention; Figure 10 Schematic diagram of the stimulation sites calculated after the arm is rotated in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0015] The present invention proposes a method for dynamic selection of intelligent electrical stimulation sites based on electrode arrays, such as Figure 1 As shown, the method aims to reconstruct basic movements such as wrist extension and flexion under a series of forearm posture changes, which are closely related to daily life activities. The method includes a surface FES subsystem for automatic stimulation site search and a PC-based human-computer interaction (HCI) feedback system. A random forest model is deployed on the sFES system for automatic stimulation site search. The model can analyze the controller's wrist movement angle data captured by the Leap Motion-based PC HCI feedback system in real time to determine the new stimulation site on the electrode array of the subject's controlled target limb. Subsequently, the system delivers electrical stimulation pulses to the newly determined stimulation site, thereby ensuring that the functional movement task performed by the controlled forearm in the previous forearm posture is not interrupted by the change of forearm posture; the method specifically includes the following steps:

[0016] S1. PC human-computer interaction feedback system developed based on Visual Studio Winform technology.

[0017] The PC-based human-computer interaction feedback system uses a Leap Motion device to capture wrist motion angle data in real time, visualizing it as numbers and waveforms. This data is then transmitted to the automatic point-finding surface FES subsystem via Wi-Fi. Furthermore, the PC host system allows users to set the electrical stimulation parameters output by the automatic point-finding surface FES subsystem and specify the stimulation site / points through the interface function module.

[0018] S2. The automatic point-finding surface FES subsystem receives wrist joint motion angle data, electrical stimulation parameter information, site / stimulation site information, and instructions such as "traverse array" and "site identification" through the ESP8266 module. "Traverse array" involves applying current pulses with fixed electrical stimulation parameters to each electrode site in the electrode array in sequence to determine whether the current electrode site can induce muscle contraction and limb movement or whether it causes pain, tingling, or other discomfort. "Site identification" involves processing the target motion angle data obtained under varying limb postures as input through a machine learning model deployed on the system to determine the new stimulation site for electrical stimulation. The received wrist joint motion angle data serves as the input signal to the machine learning model deployed on the system, and the model outputs the new effective stimulation site for electrical stimulation. The system dynamically adjusts the array electrode stimulation site for electrical stimulation by changing the wrist motion angle data input to the machine learning model in real time to ensure that the wrist movement induced by electrical stimulation remains stable during arm rotation. At the same time, the automatic point-finding body surface FES subsystem sends the currently determined stimulation site number to the PC human-computer interaction feedback system through the ESP8266 module to provide instant feedback to the user.

[0019] The system's main control MCU processes instructions received from the ESP8266 wireless WiFi module, including electrical stimulation parameters, site / stimulation site information, and wrist joint motion angle information. It then drives the digital-to-analog converter (DAC) and stimulation pulse output driver circuit to deliver corresponding positive and negative biphasic electrical stimulation pulses to the target / stimulation site. If the patient's arm position changes while performing functional movements, the MCU uses the wrist joint motion angle data as input to the machine learning model deployed on the system, intelligently determining the next valid stimulation site for outputting the electrical stimulation pulse.

[0020] Based on the array stimulation site information output by the machine learning model, the system controls the array electrode site driving circuit to close the switch of the channel corresponding to the stimulation site, so that the stimulation current is transmitted to the selected stimulation site, thereby inducing the target muscle under the site to contract and produce limb movement.

[0021] The system contains 20 stimulation channels, consisting of two 2×5 array electrodes, and each electrode in the electrode array measures 20mm×20mm.

[0022] To achieve fast and efficient data transmission between the automatic point-finding body surface FES system and the PC human-computer interaction feedback host system, the system uses the MCU's Direct Memory Access (DMA) function in conjunction with the ESP8266 wireless WiFi module to send and receive data.

[0023] The ESP8266 has complete and self-contained Wi-Fi networking capabilities. In standalone applications, the ESP8266 can boot directly from an external Flash memory and function as a Wi-Fi adapter via the serial port, supporting IEEE802.11 b / g / n protocols and a complete TCP / IP stack.

[0024] The system program adopts the STM32CubeIDE integrated development environment and uses C language for program design to realize data transmission and reception, data analysis, intelligent switching of stimulation sites, and electrical stimulation waveform generation and output.

[0025] The automatic point-finding FES subsystem uses the FreeRTOS operating system and creates four tasks: defaultTask, oled_show_task, wifi_processing_task, and update_site_task to ensure the stability of the electrical stimulation process and the overall stability and reliability of the system. These four functional tasks are as follows:

[0026] (1) defaultTask is responsible for performing necessary initialization operations after the system is reset and deleting itself after completion to ensure the reasonable allocation of system resources.

[0027] (2) oled_show_task runs when the system is idle and is responsible for updating the content of the OLED display for system debugging.

[0028] (3) wifi_processing_task is responsible for processing the stimulation parameters, stimulation sites, captured angle data, and start / stop stimulation commands received by the system through ESP8266 Wi-Fi, ensuring that the system can respond to external commands in a timely manner.

[0029] (4) update_site_task is responsible for automatically switching sites when executing the "traverse array" and "site identification" instructions.

[0030] Figure 2This section describes how the automatic point-finding surface FES subsystem responds to commands sent by the PC human-computer interaction feedback host system to generate appropriate stimulation waveforms and perform effective stimulation operations. After the system is powered on, it first executes the defaultTask function to initialize global variables. The system then enters a command listening state, awaiting commands from the PC host system. When the system receives data through the ESP8266 module, the wifi_processing_task task first verifies and parses the data, extracting electrical stimulation parameters and / or stimulation sites. It then determines whether to traverse the array sites. If not, the system drives the hardware to output electrical stimulation to the identified stimulation sites. If the received data is not a valid command, the system continues to listen.

[0031] When it's time to traverse and determine whether sites (1-20) in the electrode array are valid stimulation sites, the system executes the update_site_task task. Based on the stimulation parameters and stimulation site information analyzed by the wifi_processing_task task, the system uses a timer interrupt routine to drive the DAC and stimulation driver circuit to generate corresponding stimulation pulses, which are then transmitted to the electrode array sites to induce limb movement.

[0032] The automatic point-finding surface FES system uses a timer idle interrupt to process data received from the PC host system via the ESP8266 Wi-Fi module. If no data is transmitted from the serial port for a period of time, reception is considered complete. Therefore, there is no need to specify a length or end-of-frame marker when sending data. Table 1 shows the data received by the array electrode surface FES system in different frame formats and their definitions.

[0033] Table 1 ESP8266 model receiving and sending data frame format Changes in limb posture require the array electrode surface FES system to switch to the appropriate stimulation site in a timely and intelligent manner to ensure the coherent execution of functional rehabilitation training movements. Figure 3 Shown is a flow chart of the system's intelligent switching of stimulation sites under different body positions.

[0034] After Leap Motion is successfully initialized, the PC host computer will enter a standby state. When the user clicks the "Site Identification" button, the human-computer interaction feedback PC host computer system will immediately respond and send a command named "Site Identification" and the collected wrist motion angle data to the automatic point-finding surface FES system. After the automatic point-finding surface FES system receives and parses the command, it will perform "Site Identification" related operations. During this process, the system will use the wrist motion angle data obtained under limb posture changes as input for processing by the RF model deployed on the system to determine the new electrical stimulation site. The system will then return the determined stimulation site number to the human-computer interaction feedback PC host computer and display it as feedback in the array electrode module.

[0035] The PC human-computer interaction feedback host computer system designed based on WinForm technology in Visual Studio development environment is as follows Figure 4 As shown, it includes network settings, data transmission and reception, FES function configuration, waveform display and electrode array. The specific functions are as follows: (1) Network settings: Obtain the PC host system IP address and local port. Create a socket "server" and continuously listen for any electrode array surface FES "client" attempting to connect.

[0036] (2) Data transmission and reception: Provides visualization of data reception and transmission. Whenever data is received or sent, this area will display the sending device, sending time, and sending content.

[0037] (3) FES function configuration: Users can adjust electrical stimulation parameters such as pulse width, frequency, and current, and send instructions to the automatic point-finding body surface FES system by clicking buttons such as "Traverse Array," "Site Identification," "Start Stimulation," and "Stop Stimulation," as shown in Table 1. In addition, this module digitally displays the left and right wrist angle data obtained by Leap Motion.

[0038] (4) Waveform display module: Visual waveform display of the left and right wrist motion angle data obtained by Leap Motion.

[0039] (5) Electrode array module: allows the user to specify a "site" to have the corresponding site on the lower computer electrically stimulated. In addition, when executing the "site traversal" and "site identification" instructions, the currently selected array electrode site is displayed by displaying red visual feedback on the "site".

[0040] like Figure 5As shown in the figure, the PC host computer human-computer interaction feedback system creates a "server" and waits for the "client" of the automatic point-finding body surface FES to connect. When the "client" successfully establishes a connection with the "server", the PC host computer system initializes the Leap Motion device through the USB interface. If the Leap Motion device is successfully initialized, the host computer enters a command sending and receiving loop, continuously listening for new commands. When a new command is issued, the host computer system briefly enters a waiting state to ensure that the command is sent accurately and stably, while preparing to receive any possible responses. When the PC human-computer interaction feedback host computer receives a command from the client, it first performs data frame format verification and data parsing on the received data, and then performs corresponding operations based on the parsing results.

[0041] Given the drift issues associated with kinematic sensors like the MPU6050 when acquiring limb joint angle data, this paper employs the Leap Motion somatosensory interface to capture wrist joint angle data as arm posture changes. Leap Motion utilizes binocular cameras and stereo vision principles to construct a three-dimensional hand model, and uses a grayscale camera to reduce data volume and improve algorithm speed. These optimizations enable Leap Motion to capture hand data at rates up to 200 frames per second with an accuracy of 0.01mm, enabling real-time, fast, and accurate hand tracking, making it suitable for wrist angle data collection.

[0042] When a hand enters the recognition area, Leap Motion automatically tracks and outputs a real-time updated data frame containing information about the hand, fingers, endpoint objects, tools, gestures, and their position, velocity, direction, and rotation angle. This invention uses only wrist motion angle data, primarily relying on Leap Motion to obtain hand-related position and direction information.

[0043] Figure 6 This section describes the process of acquiring wrist motion data using the Leap Motion. After the PC (human-computer interaction feedback host computer "server") establishes a connection with the automatic point-finding surface FES "client," the PC (human-computer interaction feedback host computer) enters a state where it waits for data to be received. After capturing the data, the system graphically processes it and presents it to the user for easy observation and analysis. If data needs to be sent to a slave computer for processing, the system executes the send operation to ensure accurate transmission. Users can also save the data in .txt format for offline analysis. When no operation is required, the Leap Motion remains on standby, ready to receive new wrist motion angle data.

[0044] S3. Dynamic intelligent selection algorithm for stimulation sites.

[0045] This invention utilizes a proprietary electrode array surface functional electrical stimulation (FES) system with a monopolar electrode configuration (the stimulating electrode is each electrode site in the electrode array, and the reference electrode is a hydrogel electrode placed at the joint with a much larger area than the array electrode sites) to deliver electrical stimulation, precisely and selectively activating forearm muscles. The system includes a biphasic current pulse train that generates constant parameters, with a maximum stimulation current of 25mA. It is equipped with a pair of 2×5 array electrodes (each site is 2cm×2cm), a reference electrode (4cm×4cm), a Leap Motion somatosensory interface, a computer, and a router.

[0046] A hydrogel array electrode is applied to the dorsal and ventral sides of the arm. The system applies electrical stimulation to the site and collects wrist extension / flexion movement trajectory data under stimulation at different sites. Considering that the size and position of the electrodes in the electrode array are fixed while the size of the arm extensor and flexor muscles varies with the length of the arm, the electrodes are applied with reference to Figure 7 (a) shows the muscle distribution diagram. The electrodes are placed in the middle of the extensor carpi radialis longus and the flexor carpi ulnaris. The approximate position is as follows: Figure 7 (b) This application method is designed to effectively capture stimulation data from wrist extension and flexion movements, providing a basis for experimental analysis.

[0047] The present invention selects a machine learning random forest (RF) classification model and deploys it in the automatic point-finding surface FES subsystem to achieve dynamic selection and switching of stimulation sites under different limb postures.

[0048] A dataset of stimulation site and wrist joint motion angles across individuals and limb postures was constructed to train and evaluate the RF model. 70% of the original data was used as training data to build and adjust the model, while the remaining 30% was used as test data to evaluate the model's final predictive performance. When using the RF algorithm to build a classification model, parameter selection directly affects model complexity, fit, and classification performance. Without adjusting parameters, a cross-validation method was used to randomly shuffle the original data and divide it into ten equal parts. Then, each part of the data was used as the test set in turn, with the remaining parts used as the training set, for model training and prediction. After ten rounds of validation, the average model accuracy was calculated to be 97.2%. The training process involved: When building a classification model using the random forest algorithm, the initial performance of the model was first evaluated using cross-validation without parameter adjustment. Specifically, a ten-fold cross-validation method was used to randomly shuffle the data and divide it into ten equal parts. Model training and prediction were then performed by rotating each part of the data as the test set and the remaining parts as the training set. After ten rounds of verification, the average accuracy of the model was calculated to be 0.972515, which shows that even without parameter adjustment, the random forest model can show good performance on the data. In order to optimize the performance of the model, the n_estimators parameter was adjusted. Generally, as the number of decision trees increases, the classification accuracy of the model will increase accordingly. However, when the number of decision trees reaches a certain level, the accuracy of the model may no longer increase significantly, and may even fluctuate. In addition, too many decision trees will increase computational complexity and memory consumption, resulting in longer training time. In order to determine the appropriate range of n_estimators parameters, a learning curve is used for observation. Specifically, the value range of n_estimators is set to 1 to 150, and it increases in steps of 10. By drawing the learning curve, you can intuitively observe the changing trend of the model accuracy with the increase of n_estimators. As Figure 8 As shown in (a), when n_estimators is 61, the prediction accuracy is the highest, reaching 0.973758. In order to determine the optimal value more accurately, the learning curve is further refined in the interval [55,65], as shown in Figure 8 (b) The refined curve shows that the optimal value is still 61. Therefore, the optimal value for n_estimators is determined to be 61. Furthermore, after adjusting n_estimators, the model accuracy improved by 0.001243 compared to the previous value. The training dataset was input into the parameter-tuned random forest model. Bootstrap sampling was used to generate multiple decision trees. Each tree randomly selected a subset of features for node splitting to reduce overfitting. This completed the training of the random forest model, and the model weights and structure were saved.

[0049] To achieve intelligent switching of array electrode stimulation sites, the dynamically selected optimal classification model RF needs to be deployed to the automatic point-finding surface FES system. This paper uses the STM32CubeIDE development tool and its built-in X-CUBE-AI expansion package to parse the Open Neural Network Exchange (ONNX) data format, quickly converting the trained machine learning model into a .c file suitable for embedded systems. After the hardware deployment is complete, the model's input and output logic is written. The PC human-computer interaction feedback system program is optimized to ensure that the wrist joint motion angle data captured in real time by the Leap Motion is stably sent to the automatic point-finding surface FES hardware system at a rate of 100ms. The automatic point-finding surface FES system receives and parses the data and uses it as model input. After the model calculation and processing, the corresponding stimulation site value is output and electrical stimulation is applied.

[0050] Experimental analysis: like Figure 8 As shown in (a) and (b), when n_estimators is set to 61, the prediction accuracy is the highest, reaching 97.4%. To determine the optimal value, the learning curve is further refined in the interval [55, 65], and the optimal value is still 61.

[0051] After determining all relevant parameters, a classification model based on RF is constructed and trained using the training dataset. After training, the resulting model is used to predict test data and generate classification results. Random forest, an ensemble learning algorithm, combines the predictions of multiple decision trees to improve overall classification accuracy and stability.

[0052] Figure 9 The wrist motion curves for both the controller and the controlled were recorded, with the wrist pitch angle plotted on the ordinate and time plotted on the abscissa. The controlled wrist motion curves were similar to those of the controller, but there was a 2-3 second delay in the target movement between the controlled and the controlled. This delay is primarily due to wireless communication involving data packaging, transmission, and unpacking, which introduces communication delays. It also stems from the inability of the limb to immediately initiate movement when the stimulation site switches rapidly.

[0053] Figure 10The figure shows the effects of wrist flexion performed by a subject in three positions: pronation, neutral, and supination. The stimulation sites (red) are intelligently output by the machine learning model deployed on the system. Because the system calculates site values ​​multiple times based on captured wrist angle data during rotation, and most values ​​change instantaneously, only the site values ​​after achieving stable stimulation are shown here. There is a delay of approximately 2-3 seconds in switching stimulation sites. As can be seen from the figure, as the forearm position changes, the system is able to recalculate an appropriate stimulation site. The system outputs electrical stimulation pulses to this stimulation site, maintaining the wrist movement generated by electrical stimulation in the previous forearm position. Therefore, the experiment verifies that, by applying electrical stimulation to the new, intelligently calculated stimulation sites on the array electrode during position changes, the system can address the issue of target movement interruption caused by changes in the relative position of the surface electrodes and muscles.

[0054] As shown in Table 3, the average success rate of movement reconstruction for the six subjects reached 80.8±7.4%. This result fully demonstrates the reliability of the dynamic selection system for stimulation sites of surface FES functional electrical stimulation using an electrode array in reconstructing movements during forearm position changes. With the forearm in the neutral position, the success rate for wrist flexion reconstruction exceeded 86%, and overall, wrist flexion reconstruction was superior to wrist extension in all three forearm positions. This may be due to the fact that the same stimulation current was used to stimulate the array stimulation sites during both wrist extension and flexion. However, actual testing revealed that the stimulation current amplitude required to induce wrist extension in the dorsal forearm muscles was higher. Current amplitude is positively correlated with increased motor unit (MU) activation in muscle fibers and force generation and cross-sectional area (CSA) of muscle activation. As current intensity increases, the depolarization of MUs near the electrodes increases, leading to a gradual increase in force generation. Higher current intensities of NMES activate more muscle fibers, producing stronger muscle contractions. Therefore, the lower current intensity has a certain impact on the overall success rate. Because a fixed-size electrode array is used, even if the same stimulation site is calculated, the stimulation effect can still vary between individuals. This variability is the reason for the significant differences in movement reconstruction accuracy when different individuals perform the same target movement.

[0055] Table 3 Statistics of success rates of six groups of movements This invention uses a LeapMotion-based array electrode surface FES system (including an automatic point-finding surface FES subsystem and a PC-based human-computer interaction feedback system) to establish a stimulation site-motion angle dataset for a specific target action, which is used for algorithm model training. The intelligent algorithm model deployed on the system dynamically calculates and determines the next appropriate stimulation site that can produce the same action type based on changes in limb posture. This enables paralyzed patients to perform rehabilitation training movements that align with the rich and varied limb posture changes and sustained movement requirements in real life.

[0056] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for dynamic selection of intelligent electrical stimulation sites based on an electrode array, characterized in that: The following steps are involved: Collecting wrist joint motion angle data and corresponding stimulation site information when the user makes limb movements; after applying an electrode array to the back and ventral sides of the user's arm, applying current pulses with fixed electrical stimulation parameters to each electrode site in the electrode array in sequence to determine whether the current electrode site can induce muscle contraction to produce limb movement; A training data set is established based on the movement angle data of the wrist joint and the stimulation site information data under any specific limb movement of the user, and the random forest model is trained using the training data set; The wrist joint motion angle data of the current user to be tested is input into the trained random forest model to obtain the next effective stimulation site to output the electrical stimulation pulse; based on the effective stimulation site, the corresponding electrical stimulation pulse is generated and transmitted to the electrode site of the electrode array, thereby inducing the user's muscle contraction to produce limb movement.

2. According to the method for dynamic selection of intelligent electrical stimulation sites based on an electrode array according to claim 1, a monopolar electrode configuration electrode array surface functional electrical stimulation system is used to apply electrical stimulation to the stimulation site to obtain the motion angle data of the user's wrist joint; the monopolar electrode configuration electrode array surface functional electrical stimulation system specifically includes a stimulation electrode, a reference electrode, a biphasic current pulse sequence capable of generating constant parameters, and a Leap Motion somatosensory interactor; wherein, The stimulation electrode is each electrode site in the electrode array, and the reference electrode is an electrode placed at the joint and has an area larger than the electrode site. The Leap Motion somatosensory interactor is used to capture the movement angle data of the wrist joint under changes in arm posture after applying an electrode array to the back and ventral side of the user's arm.

3. According to the method for dynamic selection of intelligent electrical stimulation sites based on electrode arrays in claim 2, the electrode array surface functional electrical stimulation system with a monopolar electrode configuration comprises two 2×5 array electrodes and 20 stimulation channels; each electrode in the array electrode has a size of 20 mm×20 mm.

4. According to the method for dynamic selection of intelligent electrical stimulation sites based on an electrode array as described in claim 2, the Leap Motion somatosensory interactor uses binocular cameras and binocular stereo vision principles to build a three-dimensional hand model, and tracks and captures wrist joint motion angle data through the three-dimensional hand model.

5. The method for dynamic selection of intelligent electrical stimulation sites based on an electrode array according to claim 1, characterized in that: The digital-to-analog converter DAC and the stimulation pulse output driving circuit are driven by the timer interrupt program to generate corresponding electrical stimulation pulses and transmit them to the array electrode sites.

6. The method for dynamic selection of intelligent electrical stimulation sites based on an electrode array according to claim 1, characterized in that: The random forest model is trained using the training data set, specifically comprising the following steps: The training data set is randomly shuffled and divided into ten equal parts by using the ten-fold cross-validation method. Each part of the data is used as the test set in turn, and the rest are used as the training set for random forest model prediction. Tune key parameters based on the average prediction accuracy of the random forest model; The training data set is input into the random forest model after parameter tuning, and multiple decision trees are generated through Bootstrap sampling. Each tree randomly selects a feature subset for node splitting to obtain the trained random forest model.

Images