Hybrid sEMG decomposition method integrating gradient convolution kernel compensation and correlation-constrained FastICA.

Experimental testing used 5 healthy subjects, not patients, so it does not show what the method would find in people with motor disorders. The experimental signals were preprocessed data from an earlier study, not collected by this paper. Validation covered contraction levels up to 50% of maximum voluntary contraction, so performance at ultra-high forces is unknown. At low noise levels, the proposed method found very few motor units; in one simulated condition at 10 dB it found 5 ± 0 units. The refinement step adds computational time, and the method is slower than gCKC and FastICA-CKC. The method depends on the gCKC initial estimate being reliable, so periodic interference or poor initial firing sequences could limit correction.

Do not read this as evidence that the algorithm improves patient care or prosthetic control. It was tested on simulated signals and recordings from healthy subjects, not on patient outcomes.