Continuous myoelectric control of prosthetic and robotic hands requires robust regression of functional motion from surface electromyographic (sEMG) signals. However, sEMG recordings are affected by measurement noise, cross-talk, and overlapping muscle activity, which can degrade factorization-based decoding methods such as Non-Negative Matrix Factorization (NMF). We propose a hybrid approach that integrates the Frisch scheme, an Errors-in-Variables noise-consistent estimation method, with NMF-based muscle synergy extraction. The Frisch stage estimates noise-consistent linear relations among sEMG channels prior to factorization, improving the statistical structure of the data entering the NMF model. The resulting framework enables regression of continuous hand motion activation signals suitable for proportional control. Standard NMF, sparse NMF with multiplicative updates (NMF-MU), sparse NMF with alternating least squares (NMF-ALS), and the hybrid Frisch-based approach were evaluated on experimental data from five healthy subjects performing four functional hand gestures. Results show that the proposed Frisch+NMF approach achieves significantly lower reconstruction error compared to non-hybrid methods, demonstrating that noise-aware preprocessing enhances synergy-based myoelectric regression. The proposed methodology provides a principled and interpretable strategy for improving continuous sEMG-driven artificial hand control.
Meattini et al. (Sun,) studied this question.