Workflow analysis and surgical phase recognition in minimally invasive surgery

In this paper, a new approach is described to recognize the phases of a single-port sigma resection intraoperatively, based on the position signal of the surgical instruments, the endoscopic video and an audio signal, signaling coagulations. Approaches for detecting the coagulation sounds, as well as the instruments visible in the endoscopic video using a bag of words model are detailed. The intervention phases are regarded as classes of a naive Bayes classifier. Features that differentiate intervention phases are examined. The naive Bayes classifier is extended by a dynamic feature, which includes the order of the intervention phases and their duration. First results show that in 93.2% the recognized phases are classified as true positive.