A Review of Myoelectric Control for Prosthetic Hand Manipulation

Design and In Vivo Evaluation of an Intraocular Electrode for Ciliary Muscle Biopotential Measurement in a Non-Human Primate Model of Human Accommodation

The measurement of electrical potentials in the human body is becoming increasingly important in healthcare as a valuable diagnostic parameter. In ophthalmology, while these signals are primarily used to assess retinal function, other applications, such as recording accommodation-related biopotentials from the ciliary muscle, remain poorly understood. Here, we present the development and evaluation of a novel implantable ring electrode for recording biopotentials from the ciliary muscle. Inspired by capsular tension rings, the electrode was fabricated using laser cutting, wiring, and physical vapor deposition coating. The constant impedance and weight over a simulated aging period of 391 days, demonstrated the electrode's stability. In vivo testing in non-human primates further validated the electrode's surgical handling and long-term stability, with no delamination or tissue ingrowth after 100 days of implantation. Recorded biopotentials from the ciliary muscle (up to 700 µV) exceeded amplitudes reported in the literature. While the results are promising, further research is needed to investigate the signal quality and origin as well as the correlation between these signals and ciliary muscle activity. Ultimately, this electrode will be used in an implanted device to record ciliary muscle biopotentials to control an artificial lens designed to restore accommodation in individuals with presbyopia.

Gaussian process latent variable models-ANN based method for automatic features selection and dimensionality reduction for control of EMG-driven systems

Electromyography (EMG) signals have gained significant attention due to their potential applications in prosthetics, rehabilitation, and human-computer interfaces. However, the dimensionality of EMG signal features poses challenges in achieving accurate classification and reducing computational complexity. To overcome such issues, this paper proposes a novel approach that integrates feature reduction techniques with an artificial neural network (ANN) classifier to enhance the accuracy of high-dimensional EMG classification. This approach aims to improve the classification accuracy of EMG signals while substantially reducing computational costs, offering valuable implications for all EMG-related processes on such data. The proposed methodology involves extracting time and frequency domain features from twelve channels of EMG signals, followed by dimensionality reduction using techniques such as PCA, LDA, PPCA, Lasso and GPLVM, and classification using an ANN. Our investigation revealed that LDA is not appropriate for this dataset. The dimensionality reduction models did not have any significant effect on the accuracy, but the computational cost decreased significantly. In individual comparisons, GPLVM had the shortest computational time (29 s), which was significantly less than that of all the other models (p < 0.05), with PCA following at approximately 35 s and Relief at approximately 57 s, while PPCA took approximately 69 s, and Lasso exhibited higher computational costs than all the models but lower computational costs than did the original set. Using the best-performing features, all possible sets of 2, 3, 4 and 5 features were tested, and the 5-feature set exhibited the best performance. This research demonstrates the effectiveness of dimensionality reduction and feature selection in improving the accuracy of movement recognition in myoelectric control.

Biomimetic Strategies of Slip Sensing, Perception, and Protection in Prosthetic Hand Grasp

This study develops biomimetic strategies for slip prevention in prosthetic hand grasps. The biomimetic system is driven by a novel slip sensor, followed by slip perception and preventive control. Here, we show that biologically inspired sensorimotor pathways can be restored between the prosthetic hand and users. A Ruffini endings-like slip sensor is used to detect shear forces and identify slip events directly. The slip information and grip force are encoded into a bi-state sensory coding that evokes vibration and buzz tactile sensations in subjects with transcutaneous electrical nerve stimulation (TENS). Subjects perceive slip events under various conditions based on the vibration sensation and voluntarily adjust grip force to prevent further slipping. Additionally, short-latency compensation for grip force is also implemented using a neuromorphic reflex pathway. The reflex loop includes a sensory neuron and interneurons to adjust the activations of antagonistic muscles reciprocally. The slip prevention system is tested in five able-bodied subjects and two transradial amputees with and without reflex compensation. A psychophysical test for perception reveals that the slip can be detected effectively, with a success accuracy of 96.57%. A slip protection test indicates that reflex compensation yields faster grasp adjustments than voluntary action, with a median response time of 0.30 (0.08) s, a rise time of 0.26 (0.03) s, an execution time of 0.56 (0.07) s, and a slip distance of 0.39 (0.10) cm. Prosthetic grip force is highly correlated to that of an intact hand, with a correlation coefficient of 96.85% (2.73%). These results demonstrate that it is feasible to reconstruct slip biomimetic sensorimotor pathways that provide grasp stability for prosthetic users.

Multi-branch deep learning neural network prediction model for the development of angular biosensors based on sEMG

Introduction

Human gait motion intention recognition is very important for the lower extremity exoskeleton robot to accurately synchronize and respond to the user's natural motion. And motion intention recognition is generally performed through sEMG. Deep learning neural networks perform well in dealing with high-dimensional data and nonlinear relationships such as sEMG, but different deep learning neural networks have their own advantages in dealing with different types of data. Therefore, a multi-branch deep learning neural network, which enables different neural networks to process different feature items, could achieve more accurate and efficient motion intention recognition. The purpose of this study is to 1) Establish a multi-branch deep learning neural network model to achieve accurate gait recognition and effective estimation of joint angles. 2) Quantify the performance of the multi-branch deep learning neural network model in gait recognition and joint angle prediction using sEMG.

Methodology

This study involved the collection of sEMG and plantar pressure data during walking in human subjects. Firstly, the collected signals are filtered and denoised to ensure the quality and reliability of the data. Calculate the time domain features and the frequency domain features to capture the key information of gait. Then, using the sensitivity difference of different structural neural networks to different feature data, a multi-branch deep learning neural network model is developed, in which the extracted features are used as the input of the model. The output of the model includes gait cycle and joint angle, so as to realize the accurate recognition of human gait and the effective estimation of joint angle.

Results

The results show that the proposed method has high accuracy in identifying human gait and estimating joint angles. The multi-branch neural network model successfully integrates time-domain and frequency-domain features and provides reliable prediction of gait cycle and joint angle. The highest accuracy of gait recognition is 95.42%, the lowest is 90.11%, and the average is 92.16%. The average error of joint angle estimation is 3.19.

Discussion

This study designed a human walking gait recognition and joint angle prediction model to achieve accurate human lower limb motion intention recognition.The model can be integrated into the sEMG sensor to design a angular biosensors, which can predict the human joint angle in real time.

Unravelling Influence Factors in Pattern Recognition Myoelectric Control Systems: The Impact of Limb Positions and Electrode Shifts

Pattern recognition (PR)-based myoelectric control systems can naturally provide multifunctional and intuitive control of upper limb prostheses and restore lost limb function, but understanding their robustness remains an open scientific question. This study investigates how limb positions and electrode shifts-two factors that have been suggested to cause classification deterioration-affect classifiers' performance by quantifying changes in the class distribution using each factor as a class and computing the repeatability and modified separability indices. Ten intact-limb participants took part in the study. Linear discriminant analysis (LDA) was used as the classifier. The results confirmed previous studies that limb positions and electrode shifts deteriorate classification performance (14-21% decrease) with no difference between factors (p > 0.05). When considering limb positions and electrode shifts as classes, we could classify them with an accuracy of 96.13 ± 1.44% and 65.40 ± 8.23% for single and all motions, respectively. Testing on five amputees corroborated the above findings. We have demonstrated that each factor introduces changes in the feature space that are statistically new class instances. Thus, the feature space contains two statistically classifiable clusters when the same motion is collected in two different limb positions or electrode shifts. Our results are a step forward in understanding PR schemes' challenges for myoelectric control of prostheses and further validation needs be conducted on more amputee-related datasets.

Deep Generative Replay-based Class-incremental Continual Learning in sEMG-based Pattern Recognition

Developments in neural networks and sensing technologies have increased focus on modules for surface electromyogram (sEMG)-based pattern recognition. Incremental updating of parameters based on pre-trained networks can flexibly respond to user requirements and enhance user-centered interfaces. However, updating the parameters of a pre-trained network with new class data can easily lead to catastrophic forgetting. While mitigating the phenomenon by straightforwardly replaying historical data, this approach necessitates significant memory resources, a constraint that proves often impractical in real-world applications where access to historical data is limited. To avoid this limitation and incrementally add new classes to the pre-trained network, we proposed a deep generative replay-based continual learning (CL) framework. The performance was evaluated using a public sEMG dataset under two-class-incremental learning scenario until four tasks. As a result, the proposed framework performed better than other conventional CL methods except for experience replay, which simply reuses historical data.

Enhancing Myoelectric Prosthetic Control: Deep Learning Strategies for Continuous Arm Kinematics Estimation and Cross-Subject Model Transferability from EMG Data

Over the last decade, myoelectric prosthesis control has witnessed considerable advancements, yet there remain significant challenges. Two key constraints have been the limited range of movements and the lack of simultaneous control capabilities. This study aims to address these issues by introducing an LSTM-based approach for the continuous control of critical parameters in prosthetic limbs. Utilizing deep learning models, our method enhances the precision in controlling the elbow angle (θ), as well as the horizontal (X) and vertical (Y) positions of the wrist joint, coupled with the velocity (v). This research not only focuses on the spatial and dynamic aspects of the movement but also emphasizes the transferability of our model across different subjects. We have successfully demonstrated the model's ability to be trained on one subject and applied effectively to another.

TinyML for Real-Time Embedded HD-EMG Hand Gesture Recognition with On-Device Fine-Tuning

This work introduces a fully embedded wireless platform that incorporates the Coral Tensor Processing Unit (TPU) accelerator to leverage TinyML for real-time hand gesture recognition using high-density surface electromyography (HD-sEMG). With a general inference time of 2.96 ms using a 64 channels sensor, the TPU proved to be well suited for such real-time recognition tasks. Constructed from off-the-shelf components, the platform offers a cost-effective and self-sufficient alternative for integrating artificial intelligence into prosthetic devices, eliminating the dependency on expensive external hardware. The system allows for intuitive calibration through a user interface, facilitating fine-tuning of the inference model directly on the device or remotely via a cloud-based server. On-device finetuning yields similar performance to the cloud-based approach, improving gesture recognition accuracy by up to 36.15% in intersession test cases. Extensive exploration of 8-bit data quantization techniques demonstrates that hardware compatibility can be achieved without sacrificing performance. In the best case, the proposed quantization scheme can improve the results by 0.96% compared to unquantized data. Overall, this paper establishes a robust foundation for advancing on-device HD-sEMG based hand gesture recognition, paving the way for more accessible and practical myoelectric prosthetic solutions.

Simultaneous Three-Degrees-of-Freedom Prosthetic Control Based on Linear Regression and Closed-Loop Training Protocol

Machine learning-based controllers of prostheses using electromyographic signals have become very popular in the last decade. The regression approach allows a simultaneous and proportional control of the intended movement in a more natural way than the classification approach, where the number of movements is discrete by definition. However, it is not common to find regression-based controllers working for more than two degrees of freedom at the same time. In this paper, we present the application of the adaptive linear regressor in a relatively low-dimensional feature space with only eight sensors to the problem of a simultaneous and proportional control of three degrees of freedom (left-right, up-down and open-close hand movements). We show that a key element usually overlooked in the learning process of the regressor is the training paradigm. We propose a closed-loop procedure, where the human learns how to improve the quality of the generated EMG signals, helping also to obtain a better controller. We apply it to 10 healthy and 3 limb-deficient subjects. Results show that the combination of the multidimensional targets and the open-loop training protocol significantly improve the performance, increasing the average completion rate from 53% to 65% for the most complicated case of simultaneously controlling the three degrees of freedom.

Shoulder Movement-Centered Measurement and Estimation Scheme for Underarm-Throwing Motions

Underarm throwing motions are crucial in various sports, including boccia. Unlike healthy players, people with profound weakness, spasticity, athetosis, or deformity in the upper limbs may struggle or find it difficult to control their hands to hold or release a ball using their fingers at the proper timing. To help them, our study aims to understand underarm throwing motions. We start by defining the throwing intention in terms of the launch angle of a ball, which goes hand-in-hand with the timing for releasing the ball. Then, an appropriate part of the body is determined in order to estimate ball-throwing intention based on the swinging motion. Furthermore, the geometric relationship between the movements of the body part and the release angle is investigated by involving multiple subjects. Based on the confirmed correlation, a calibration-and-estimation model that considers individual differences is proposed. The proposed model consists of calibration and estimation modules. To begin, as the calibration module is performed, individual prediction states for each subject are updated online. Then, in the estimation module, the throwing intention is estimated employing the updated prediction. To verify the effectiveness of the model, extensive experiments were conducted with seven subjects. In detail, two evaluation directions were set: (1) how many balls need to be thrown in advance to achieve sufficient accuracy; and (2) whether the model can reach sufficient accuracy despite individual differences. From the evaluation tests, by throwing 20 balls in advance, the model could account for individual differences in the throwing estimation. Consequently, the effectiveness of the model was confirmed when focusing on the movements of the shoulder in the human body during underarm throwing. In the near future, we expect the model to expand the means of supporting disabled people with ball-throwing disabilities.

Machine learning for hand pose classification from phasic and tonic EMG signals during bimanual activities in virtual reality

Myoelectric prostheses have recently shown significant promise for restoring hand function in individuals with upper limb loss or deficiencies, driven by advances in machine learning and increasingly accessible bioelectrical signal acquisition devices. Here, we first introduce and validate a novel experimental paradigm using a virtual reality headset equipped with hand-tracking capabilities to facilitate the recordings of synchronized EMG signals and hand pose estimation. Using both the phasic and tonic EMG components of data acquired through the proposed paradigm, we compare hand gesture classification pipelines based on standard signal processing features, convolutional neural networks, and covariance matrices with Riemannian geometry computed from raw or xDAWN-filtered EMG signals. We demonstrate the performance of the latter for gesture classification using EMG signals. We further hypothesize that introducing physiological knowledge in machine learning models will enhance their performances, leading to better myoelectric prosthesis control. We demonstrate the potential of this approach by using the neurophysiological integration of the "move command" to better separate the phasic and tonic components of the EMG signals, significantly improving the performance of sustained posture recognition. These results pave the way for the development of new cutting-edge machine learning techniques, likely refined by neurophysiology, that will further improve the decoding of real-time natural gestures and, ultimately, the control of myoelectric prostheses.

Does Exerting Grasps Involve a Finite Set of Muscle Patterns? A Study of Intra- and Intersubject Variability of Forearm sEMG Signals in Seven Grasp Types

Surface Electromyography (sEMG) signals are widely used as input to control robotic devices, prosthetic limbs, exoskeletons, among other devices, and provide information about someone's intention to perform a particular movement. However, the redundant action of 32 muscles in the forearm and hand means that the neuromotor system can select different combinations of muscular activities to perform the same grasp, and these combinations could differ among subjects, and even among the trials done by the same subject. In this work, 22 healthy subjects performed seven representative grasp types (the most commonly used). sEMG signals were recorded from seven representative forearm spots identified in a previous work. Intra- and intersubject variability are presented by using four sEMG characteristics: muscle activity, zero crossing, enhanced wavelength and enhanced mean absolute value. The results confirmed the presence of both intra- and intersubject variability, which evidences the existence of distinct, yet limited, muscle patterns while executing the same grasp. This work underscores the importance of utilizing diverse combinations of sEMG features or characteristics of various natures, such as time-domain or frequency-domain, and it is the first work to observe the effect of considering different muscular patterns during grasps execution. This approach is applicable for fine-tuning the control settings of current sEMG devices.

Decoding Multi-DoF Movements Using a CST-Based Force Generation Model With Single-DoF Training

Recent developments in dexterous myoelectric prosthetics have established a hardware base for human-machine interfaces. Although pattern recognition techniques have seen successful deployment in gesture classification, their applications remain largely confined to certain specific discrete gestures. Addressing complex daily tasks demands an immediate need for precise simultaneous and proportional control (SPC) for multiple degrees of freedom (DoFs) movements. In this paper, we introduce an SPC approach for multi-DoF wrist movements using the cumulative spike trains (CSTs) of motor unit pools, merely leveraging single-DoF training. The efficacy of our proposed approach was validated offline against existing methods respectively based on non-negative matrix factorization and motor unit spike trains, using experimental data. The experimental process includes both single-DoF (for training) and multi-DoF (for testing) movements. We evaluated the performance using Pearson correlation coefficient (R) and the normalized root mean square error (nRMSE). The results reveal that our method outperforms comparative approaches in force estimation for both testing datasets (3 and 4). On average, for dataset 3, R and nRMSE of the flexion/extension DoF (the pronation/supination DoF) are 0.923±0.037 (0.901±0.040) and 12.3±3.1% (12.9±2.2%); similarly, those of dataset 4 are 0.865±0.057 (0.837±0.053) and 14.9±2.9% (15.4±2.0%), respectively. The outcomes demonstrate the effectiveness of our method in simultaneous and proportional force estimation for multi-DoF wrist movements, showing a promising potential as a neural-machine interface for SPC of dexterous myoelectric prostheses.