PCA-based muscle selection for interventional manipulation recognition

Abstract

Vascular interventional robot is becoming increasingly popular in assisting doctors for the treatment of cardiovascular diseases (CVDs). However, natural manipulations and motion patterns of surgeons in percutaneous coronary intervention (PCI), including finger motion and hand motion, are potentially altered more or less through this solution. Since clinical success is highly dependent on the skills and dexterous manipulation strategies of the surgeons, but these skills and experience remains an underused resource in robot-assisted intervention. In this paper, surface electromyography (sEMG) of surgeons' hand and arm muscles, a feature of manipulation skills in intervention, is acquired to recognize six interventional manipulations with Back-Propagation (BP) neural network. A modified principal component analysis (PCA) is presented to select a sensitive and principal muscle subset for improving the fiexibility of subjects' manipulations. Experimental results show that the sensitive muscle subset with the recognition rate over 90% performs better than the insensitive one, and the sEMG-based BP neural network has the ability to recognize interventional manipulations with a high accuracy.