Towards a Feasible Hand Gesture Recognition System as Sterile Non-contact Interface in the Operating Room with 3D Convolutional Neural Network

Abstract

Operating surgeons are constrained when interacting with computer systems as they traditionally utilize hand-held devices such as keyboard and mouse. Studies have previously proposed and shown the use of hand gestures is an efficient, touchless way of interfacing with such systems to maintain a sterile field. In this paper, we propose a Deep Computer Vision-based Hand Gesture Recognition framework to facilitate the interaction. We trained a 3D Convolutional Neural Network with a very large scale dataset to classify hand gestures robustly. This network became the core component of a prototype application requiring intraoperative navigation of medical images of a patient. Usability evaluation with surgeons demonstrates the application would work and a hand gesture lexicon that is germane to Medical Image Navigation was defined. By completing one cycle of usability engineering, we prove the feasibility of using the proposed framework inside the Operating Room.