A modified ICA framework for motion artifact removal in wrist-type photoplethysmography during exercise

Removal of motion artifacts (MA) from wrist-type photoplethysmographic (PPG) signal recordings during exercise is a difficult problem, since the MA during exercise can be very strong. In this paper, a modified independent component analysis (ICA) algorithm to remove MA is proposed. The proposed algorithm relies on the negentropy scores of the linear combinations of the signal data to achieve maximum statistical independence. This algorithm was tested on PPG signals recorded during fast running. The results of the experiments reveal that this algorithm is robust to MA present in the PPG signals. The results of our algorithm are compared with the existing wavelet-decomposition method.