Spatial-temporal Modelling for Surgical Tool Classification in Cholecystectomy Videos

Abstract

Surgical tool classification is an essential component to analyse the surgical workflow of laparoscopic intervention. It has many potential applications, for instance, developing decision-support systems, automatic indexing of laparoscopic videos, and assessing surgical skills. In this work, a framework for surgical tool presence detection is presented. The proposed approach consists of a CNN model and two LSTM units to model spatial and temporal information encoded in the laparoscopic video. The proposed approach achieved a mean average precision of 94.57%. Experimental results show the value of temporal modelling in improving the classification performance of surgical tools.