Pattern recognition of electromyography (EMG) signals can potentially improve the performance of myoelectric control for upper limb prostheses with respect to current clinical approaches based on direct control. However, the choice of features for classification is challenging and impacts long-term performance. Here, we propose the use of EMG raw signals as direct inputs to deep networks with intrinsic feature extraction capabilities recorded over multiple days. Seven able-bodied subjects performed six active motions (plus rest), and EMG signals were recorded for 15 consecutive days with two sessions per day using the MYO armband (MYB, a wearable EMG sensor). The classification was performed by a convolutional neural network (CNN) with raw bipolar EMG samples as the inputs, and the performance was compared with linear discriminant analysis (LDA) and stacked sparse autoencoders with features (SSAE-f) and raw samples (SSAE-r) as inputs. CNN outperformed (lower classification error) both LDA and SSAE-r in the within-session, between sessions on same day, between the pair of days, and leave-out one-day evaluation (p < 0.001) analyses. However, no significant difference was found between CNN and SSAE-f. These results demonstrated that CNN significantly improved performance and increased robustness over time compared with standard LDA with associated handcrafted features. This data-driven features extraction approach may overcome the problem of the feature calibration and selection in myoelectric control.

This study describes the first application of a support vector machine (SVM) based scheme for real-time simultaneous and proportional myoelectric control of multiple degrees of freedom (DOFs). Three DOFs including wrist flexion-extension, abduction-adduction and forearm pronation-supination were investigated with 10 able-bodied subjects and two individuals with transradial limb deficiency (LD). A Fitts' law test involving real-time target acquisition tasks was conducted to compare the usability of the SVM-based control system to that of an artificial neural network (ANN) based method. Performance was assessed using the Fitts' law throughput value as well as additional metrics including completion rate, path efficiency and overshoot. The SVM-based approach outperformed the ANN-based system in every performance measure for able-bodied subjects. The SVM outperformed the ANN in path efficiency and throughput with the first LD subject and in throughput with the second LD subject. The superior performance of the SVM-based system appears to be due to its higher estimation accuracy of all DOFs during inactive and low amplitude segments (these periods were frequent during real-time control). Another advantage of the SVM-based method was that it substantially reduced the processing time for both training and real time control.

Brain-computer interfaces can be used for motor substitution and recovery; therefore, detection and classification of movement intention is crucial for optimal control. In this study, palmar, lateral and pinch grasps were differentiated from the idle state and classified from single-trial EEG using only information prior the movement onset. Fourteen healthy subjects performed the three grasps 100 times while EEG was recorded from 25 electrodes. Temporal and spectral features were extracted from each electrode, and feature reduction was performed using sequential forward selection (SFS) and principal component analysis (PCA).The detection problem was investigated as the ability to discriminate between movement preparation and the idle state. Furthermore, all task pairs and the three movements together were classified. The best detection performance across movements (79±8%) was obtained by combining temporal and spectral features. The best movement-movement discrimination was obtained using spectral features; 76±9% (2-class) and 63±10% (3-class). For movement detection and discrimination, the performance was similar across grasp types and task pairs; SFS outperformed PCA. The results show it is feasible to detect different grasps and classify the distinct movements using only information prior to the movement onset; which may enable brain-computer interface-based neurorehabilitation of upper limb function through Hebbian learning mechanisms.

BackgroundIn this paper, we propose a nonlinear minimally supervised method based on autoencoding (AEN) of EMG for myocontrol. The proposed method was tested against the state-of-the-art (SOA) control scheme using a Fitts’ law approach.MethodsSeven able-bodied subjects performed a series of target acquisition myoelectric control tasks using the AEN and SOA algorithms for controlling two degrees-of-freedom (radial/ulnar deviation and flexion/extension of the wrist), and their online performance was characterized by six metrics.ResultsBoth methods allowed a completion rate close to 100%, however AEN outperformed SOA for all other performance metrics, e.g. it allowed to perform the tasks on average in half the time with respect to SOA. Moreover, the amount of information transferred by the proposed method in bit/s was nearly twice the throughput of SOA.ConclusionsThese results show that autoencoders can map EMG signals into kinematics with the potential of providing intuitive and dexterous control of artificial limbs for amputees.

Currently, most of the adopted myoelectric schemes for upper limb prostheses do not provide users with intuitive control. Higher accuracies have been reported using different classification algorithms but investigation on the reliability over time for these methods is very limited. In this study, we compared for the first time the longitudinal performance of selected state-of-the-art techniques for Electromyography (EMG) based classification of hand motions. Experiments were conducted on ten able-bodied and six transradial amputees for seven continuous days. Linear Discriminant Analysis (LDA), Artificial Neural Network (ANN), Support Vector Machine (SVM), K-Nearest Neighbour (KNN) and Decision Trees (TREE) were compared. Comparative analysis showed that the ANN attained highest classification accuracy followed by LDA. Three-way repeated ANOVA test showed a significant difference ($\rm P\lt 0.001$) between EMG types (surface, intramuscular and combined), Days (1-7), classifiers and their interactions. Performance on last day was significantly better ($\rm P\lt 0.05$) than the first day for all classifiers and EMG types. Within-day classification error (WCE) across all subject and days in ANN was: surface (9.12 ± 7.38%), intramuscular (11.86±7.84%) and combined (6.11±7.46%). The between-day analysis in a leave-one-day-out fashion showed that ANN was the optimal classifier (surface (21.88 ± 4.14%) intramuscular (29.33 ± 2.58%) and combined (14.37 ± 3.10%)). Results indicate that that within day performances of classifiers may be similar but over time it may lead to a substantially different outcome. Furthermore, training ANN on multiple days might allow capturing time-dependent variability in the EMG signals and thus minimizing the necessity for daily system recalibration.