High-density Surface EMG Signal Quality Enhancement Via Optimized Filtering Technique for Amputees’ Motion Intent Characterization Towards Intuitive Prostheses Control

Electromyogram (EMG) based pattern recognition (PR) strategies have been well investigated and applied as viable control strategies for upper limb prostheses. Transhumeral amputees often do not have enough residual muscles to produce high-quality signals needed to control the device adequately, so EMG-PR based prostheses for above-elbow amputees have not yet gained widespread acceptance in clinical and commercial settings. The limited acquired signal from these amputees often contains noises that make it challenging to accurately decode their limb movement intent. Thus, this study explored the capability of optimized filtering techniques (including Wiener, Hampel, and 1-Dimensional Median filters) in denoising the EMG signal towards improving its overall quality to realize adequate EMG-PR based control strategies. The performance of the filtering techniques was investigated using myoelectric signals collected from transhumeral amputees. Five different time-domain features and two commonly adopted classifiers were employed to classify the limb movement tasks, while metrics including Accuracy, Recall, and F-score were used for evaluation. Experimental results indicate that the Wiener filter (WF) achieved an increment of up to 6.24%, 6.38%, and 6.95% in accuracy compared to the other filtering methods using the original data, data with random noise, and data with white Gaussian noise, respectively. Also, T-Test statistical analysis results show that WF led to improved performance with statistical significance (p < 0.05) over all other considered methods across metrics and classifiers, with an improvement of up to 18.64% for individual classes of motions in terms of accuracy. Findings from this study indicate that the proposed method enhanced the amputees' EMG signal quality which led to substantial improvement in classification of their limb movement intents and the method may positively impact the EMG-PR based prostheses control schemes.