Segmentation of Endonasal Robotic Instruments in a Head Phantom using Deep Learning and Virtual-Reality Simulation

Abstract

The manual generation of training data for the semantic segmentation of medical images using deep neural-networks is a time-consuming and error-prone task. In this study, we evaluate the applicability of realistic synthetic images in the training of deep-learning models. We used virtual reality (VR) simulation to generate photorealistic images and ground-truth segmentation of endonasal robotic instruments in a head phantom. A convolutional neural network called U-Net was trained using the synthetic databases and validated on real images.