Elbow-flexion force estimation during arm posture dynamically changing between pronation and supination

Neuro-Musculoskeletal Modeling for Online Estimation of Continuous Wrist Movements from Motor Unit Activities

Decoding movement intentions from motor unit (MU) activities remains an ongoing challenge, which restricts our comprehension of the intricate transition mechanism from microscopic neural drive to macroscopic movements. This study presents an innovative neuro-musculoskeletal (NMS) model driven by MU activities for online estimation of continuous wrist movements. The proposed model employs a physiological and comprehensive utilization of MU firings and waveforms, thus facilitating the localization of MUs to muscle-tendon units (MTU) as well as the computation of MU-specific neural excitation. Subsequently, the MU-specific neural excitation was integrated to form the MTU-specific neural excitation, which were then inputted into a musculoskeletal model to accomplish the joint angle estimation. To assess the effectiveness of this model, high-density surface electromyography and angular data were collected from the forearms of eight subjects during their performance of wrist flexion-extension task. Two pieces of 8×8 electrode arrays and a motion capture system were employed for data acquisition. Following offline model calibration with a global optimization algorithm, online angle estimation results demonstrated a significant superiority of the proposed model over the state-of-the-art NMS models (p < 0.05), yielding the lowest normalized root mean square error ( 0.10 ± 0.02 ) and the highest determination coefficient ( 0.87 ± 0.06 ). This study provides a novel idea for the decoding of joint movements from MU activities. The research findings hold the potential to advance the development of NMS models towards the control of multiple degrees of freedom, with promising applications in the fields of motor control, biomechanics, and neuro-rehabilitation engineering.

An Adaptive Hammerstein Model for FES-Induced Torque Prediction Based on Variable Forgetting Factor Recursive Least Squares Algorithm

Modeling the muscle response to functional electrical stimulation (FES) is an important step during model-based FES control system design. The Hammerstein structure is widely used in simulating this nonlinear biomechanical response. However, a fixed relationship cannot cope well with the time-varying property of muscles and muscle fatigue. In this paper, we proposed an adaptive Hammerstein model to predict ankle joint torque induced by electrical stimulation, which used variable forgetting factor recursive least squares (VFFRLS) method to update the model parameters. To validate the proposed model, ten healthy individuals were recruited for short-duration FES experiments, ten for long-duration FES experiments, and three stroke patients for both. The isometric ankle dorsiflexion torque induced by FES was measured, and then the test performance of the fixed-parameter Hammerstein model, the adaptive Hammerstein model based on fixed forgetting factor recursive least squares (FFFRLS) and the adaptive Hammerstein model based on VFFRLS was compared. The goodness of fit, root mean square error, peak error and success rate were applied to evaluate the accuracy and stability of the model. The results indicate a significant improvement in both the accuracy and stability of the proposed adaptive model compared to the fixed-parameter model and the adaptive model based on FFFRLS. The proposed adaptive model enhances the ability of the model to cope with muscle changes.

MuscleNET: mapping electromyography to kinematic and dynamic biomechanical variables by machine learning

Objective.This paper proposes machine learning models for mapping surface electromyography (sEMG) signals to regression of joint angle, joint velocity, joint acceleration, joint torque, and activation torque.Approach.The regression models, collectively known as MuscleNET, take one of four forms: ANN (forward artificial neural network), RNN (recurrent neural network), CNN (convolutional neural network), and RCNN (recurrent convolutional neural network). Inspired by conventional biomechanical muscle models, delayed kinematic signals were used along with sEMG signals as the machine learning model's input; specifically, the CNN and RCNN were modeled with novel configurations for these input conditions. The models' inputs contain either raw or filtered sEMG signals, which allowed evaluation of the filtering capabilities of the models. The models were trained using human experimental data and evaluated with different individual data.Main results.Results were compared in terms of regression error (using the root-mean-square) and model computation delay. The results indicate that the RNN (with filtered sEMG signals) and RCNN (with raw sEMG signals) models, both with delayed kinematic data, can extract underlying motor control information (such as joint activation torque or joint angle) from sEMG signals in pick-and-place tasks. The CNNs and RCNNs were able to filter raw sEMG signals.Significance.All forms of MuscleNET were found to map sEMG signals within 2 ms, fast enough for real-time applications such as the control of exoskeletons or active prostheses. The RNN model with filtered sEMG and delayed kinematic signals is particularly appropriate for applications in musculoskeletal simulation and biomechatronic device control.

Optimal strategy of sEMG feature and measurement position for grasp force estimation

Grasp force estimation based on surface electromyography (sEMG) is essential for the dexterous control of a prosthetic hand. Nowadays, although increasing the number of sEMG measurement positions and extracting more features are common methods to increase the accuracy of grasp force estimation, it will increase the computational burden. In this paper, an approach based on analysis of variance (ANOVA) and generalized regression neural network (GRNN) for optimal measurement positions and features is proposed, with the purpose of using fewer measurement positions or features to achieve higher estimation accuracy. Firstly, we captured six channels of sEMG from subjects’ forearm and grasp force synchronously. Then, four kinds of features in time domain are extracted from each channel of sEMG. By combining different measurement position sets (MPSs) and feature set (FSs), we construct 945 data sets. These data sets are fed to GRNN to realize grasp force estimation. Normalized root mean square error (NRMS), normalized mean of absolute error (NMAE), and correlation coefficient (CC) between estimated grasp force and actual force are introduced to evaluate the performance of grasp force estimation. Finally, ANOVA and Tukey HSD testing are introduced to analyze grasp force estimation results so as to obtain the optimal measurement positions and features. We obtain the optimal MPSs for grasp force estimation when different FSs are employed, and the optimal FSs when different MPSs are utilized.

Adaptive Electrode Calibration Method Based on Muscle Core Activation Regions and Its Application in Myoelectric Pattern Recognition

To reduce the bad effect of electrode shifts on myoelectric pattern recognition, this paper presents an adaptive electrode calibration method based on core activation regions of muscles. In the proposed method, the high-density surface electromyography (HD-sEMG) matrix collected during hand gesture execution is decomposed into source signal matrix and mixed coefficient matrix by fast independent component analysis algorithm firstly. The mixed coefficient vector whose source signal has the largest two-norm energy is selected as the major pattern, and core activation region of muscles is extracted by traversing the major pattern periodically using a sliding window. The electrode calibration is realized by aligning the core activation regions in unsupervised way. Gestural HD-sEMG data collection experiments with known and unknown electrode shifts are carried out on 9 gestures and 11 participants. A CNN+LSTM-based network is constructed and two network training strategies are adopted for the recognition task. The experimental results demonstrate the effectiveness of the proposed method in mitigating the bad effect of electrode shifts on gesture recognition accuracy and the potentials in reducing user training burden of myoelectric control systems. With the proposed electrode calibration method, the overall gesture recognition accuracies increase about (5.72~7.69)%. In specific, the average recognition accuracy increases (13.32~17.30)% when using only one batch of data in data diversity strategy, and increases (12.01~13.75)% when using only one repetition of each gesture in model update strategy. The proposed electrode calibration algorithm can be extended and applied to improve the robustness of myoelectric control system.

Upper Limb End-Effector Force Estimation During Multi-Muscle Isometric Contraction Tasks Using HD-sEMG and Deep Belief Network

In this study, research was carried out on the end-effector force estimation of two representative multi-muscle contraction tasks: elbow flexion and palm-pressing. The aim was to ascertain whether an individual muscle or a combination of muscles is more suitable for the end-effector force estimation. High-density surface electromyography (HD-sEMG) signals were collected from four primary muscle areas of the upper arm and forearm: the biceps brachii (BB), brachialis (BR), triceps brachii (TB), brachioradialis (BRD), and extensor digitorum communis (EDC). The wrist pulling and palm-pressing forces were measured in elbow flexion and palm-pressing tasks, respectively. The deep belief network (DBN) was adopted to establish the relation between HD-sEMG and the measured force. The representative signals of the four primary areas, which were considered as the input signal of the force estimation model, were extracted by HD-sEMG using the principle component analysis (PCA) algorithm, and fed separately or together into the DBN. An index termed mean impact value (MIV) was proposed to describe the priority of different muscle groups for estimating the end-effector force. The experimental results demonstrated that, in multi-muscle isometric contraction tasks, the dominant muscles with the highest activation degree could track variations in the end-effector force more effectively, and are more suitable than a combination of muscles. The main contributions of this research are as follows: (1) To fuse the activation information from different muscles effectively, DBN was adopted to establish the relationship between HD-sEMG and the generated force, and achieved highly accurate force estimation. (2) Based on the well-trained DBN force estimation model, an index termed MIV was presented to evaluate the priority of muscles for estimating the generated force.