Evaluation of regression methods for the continuous decoding of finger movement from surface EMG and accelerometry

The reconstruction of finger movement activity from surface electromyography (sEMG) has been proposed for the proportional and simultaneous myoelectric control of multiple degrees-of-freedom (DOFs). In this paper, we propose a framework for assessing decoding performance on novel movements, that is movements not included in the training dataset. We then use our proposed framework to compare the performance of linear and kernel ridge regression for the reconstruction of finger movement from sEMG and accelerometry. Our findings provide evidence that, although the performance of the non-linear method is superior for movements seen by the decoder during the training phase, the performance of the two algorithms is comparable when generalizing to novel movements.