Adaptive EMG Pattern Recognition Reduces Frequency and Improves Quality of At-home Prosthesis Training For Upper Limb Myoelectric Prosthesis Wearers
AbstractUpper limb myoelectric pattern recognition-controlled prostheses use machine learning algorithms to identify a wearer's intended movement from their muscle activity patterns. However, many factors can contribute to changes in the characteristics of the EMG input signals (electrode shift, muscle fatigue, limb position etc.) during everyday prosthesis use which can diminish controller performance. Multiple in-lab studies have demonstrated promising results towards improving controller performance by employing advanced algorithms, none of which have been tested clinically, that can adapt to these changes. This paper presents the implementation of a supervised-adaptation algorithm on a commercially available pattern recognition control system that makes use of historical EMG data collected during previous user-initiated calibration routines to update the existing classification model. In an at-home clinical study, we evaluated whether real-world use of adaptive classification reduces how often upper limb prosthesis wearers need to recalibrate their pattern recognition system.
How to Cite
Z. Wright and B. Lock, “Adaptive EMG Pattern Recognition Reduces Frequency and Improves Quality of At-home Prosthesis Training For Upper Limb Myoelectric Prosthesis Wearers”, MEC Symposium, Aug. 2022.
Clinical Research Studies