Semantic Segmentation of Human Torso Region for minimal invasive Surgery

Abstract

We propose a deep learning based semantic segmentation algorithm to identify and label the tissues and organs in the endoscopic video feed of the human torso region. Our contributions in this project are two-fold: first, we contribute an annotated dataset created from actual endoscopic video feed of surgical procedures, and secondly, we propose a deep neural network for semantic segmentation. To cater to the low quantity of annotated data, we propose unsupervised pre-training and data augmentation. The trained model is evaluated on the independent test set of the proposed dataset. This thesis serves as the first step towards autonomous minimal invasive surgery.