Adaptive Motor Unit Decomposition During Walking: Towards Systematic Validation

Understanding motor unit (MU) behavior during locomotion is essential for uncovering the neural control of human movement and advancing the development of assistive technologies. Although high-density electromyography (RDEMG) allows for the extraction of individual MU firing activity, conventional methods are typically restricted to isometric or slow dynamic contractions. Additionally, accurate characterization of MU twitch responses is essential for decoding the mechanical output of muscle contractions. In this study, we present an adaptive online-capable methodology for decoding soleus' MU firing events and MU-specific activation dynamics along with a conservative validation during locomotion. Our proposed framework incorporates advanced blind source separation techniques, adapted for the detection of low-threshold MUs and capable of addressing non-stationary MU action potential variations. The framework is built on three main components: (1) an offline process to derive decomposition parameters and optimal twitch characteristics from isometric and walking trials, (2) an online approach that includes both static and adaptive decomposition strategies to refine MU firing event estimation and apply derived twitch properties for real-time MU-specific activation, and (3) a validation of the decoded MU spike trains against those derived from intramuscular EMG, as well as a comparison of MU-specific activation against ankle joint moments. Experimental results demonstrated that the adaptive approach improved the agreement with intramuscular EMG-derived spike trains and provided strong correlations between MU-specific activation and ankle joint moments. This work holds significant promise for future real-time applications in assistive devices that can more effectively adapt to individual patient needs and track their progress over time.