Surgical phase recognition using movement data from video imagery and location sensor data

Abstract

Abstract The automatic recognition of surgical phases has strong potential to help medical staff understand individual and group patterns, optimize work flows, and identify potential work flow risks that lead to adverse medical events in an operating room. In this chapter, we investigate the performance of context recognition on the movement of operating room staff throughout their work environment, which was measured by imaging and tracking. We employed an optical flow algorithm and trajectory clustering techniques to extract movement characteristics of surgical staff from video imagery and time-stamped location data collected by an ultrasonic location aware system, respectively. Then we applied a Support Vector Machine to time-stamped location data, optical flow estimates, trajectory clusters, and combinations of these three data to examine the intraoperative context recognition rate. Our results show that the integration of both video imagery and location sensor data improves context awareness of neurosurgical operations.