Improving the identification of finger movements using high-density surface electromyography pre-processed with PCA

Abstract

We investigated whether identification of different finger tasks only relying on the agonist or antagonist extensor digitorum communis (EDC) can be improved by using high-density sEMG (HDsEMG) pre-processed with principal component analysis (PCA). Monopolar HDsEMG was respectively recorded from EDC when the EDC muscle respectively acted as agonist or antagonist muscles. PCA-based approach was evaluated using k-nearest neighbour (KNN) classifier and compared with the classical spatial filters. Using PCA-based configuration can achieve better classification performance in identification of tasks and effort levels and dramatically outperformed spatial filtering configurations in all cases (p<0.05). It can be concluded that PCA can replace the prevailing spatial filters as a general procedure pre-processed HDsEMG, showing that distinct activation distribution patterns of EDC muscle as a function of individual finger flexion as well as extension and its corresponding contraction levels.