Longitudinal analysis of neuromuscular adaptation using entropy and PCA during motor learning

Motor learning can yield clear behavioral gains even when EMG-derived indices show subtle, heterogeneous, or non-monotonic change. We aimed to describe longitudinal trajectories of performance and multi-channel EMG summaries during extended practice of a tempo-constrained finger sequencing task under stable constraints.

Eleven participants practiced a fixed sequence for 30 sessions. Timing accuracy was quantified as mean absolute timing error. Forearm/hand EMG was normalized to a reference voluntary contraction (%RVC). We computed EMG entropy as a summary of the spatial dispersion of activation, summarized early-to-late changes in normalized amplitude (Δ%RVC), and used a pooled (common) PCA framework to define a dataset-level dominant covariation axis and track session-related changes in PC1 scores (projections) along this fixed axis. Longitudinal trends were examined using linear mixed-effects models.

Timing error decreased with practice in a pattern consistent with a negatively accelerated (quadratic) trajectory. Group-level entropy did not show a uniform monotonic trend, and individual entropy trajectories varied substantially. Δ%RVC changes were generally small and non-uniform across muscles and participants. In the pooled PCA framework, PC1 scores exhibited a modest session-related drift, suggesting gradual shifts in activation patterns along a common covariation dimension.

Under fixed task constraints, behavioral improvement may coexist with heterogeneous EMG entropy trajectories and small, non-uniform changes in normalized amplitude. Multi-channel summaries can help characterize individualized routes to skill acquisition, whereas mechanistic interpretation will likely require additional measures (e.g., kinematics/kinetics) and/or explicit manipulations of task constraints.