Deep Learning and Augmented Reality application for minimally invasive urologic surgery support

Artificial intelligence and deep learning are becoming increasingly popular in the medical field. In recent years, neural networks are consolidating as a support tool for doctors in the phases of diagnosis, prognosis, and in the operating room during delicate surgery. This thesis aims to implement an Augmented Reality application for improving the surgeon’s spatial perception during Robot-Assisted Radical Prostatectomy (RARP) stages. RARP is a minimally invasive urologic operation performed with Da Vinci surgical console support. This technology improves surgical precision, reducing post-operative complications and hospital stays. The focus is on finding a technique for real-time automatic registration. The proposed system trains a neural network to estimate the position, rotation, and scale of the object captured by the endoscope and intends to obtain an optimal overlap between the 3D virtual prostate model and its physical counterpart. This work includes an introduction about thesis aim and context, theoretical notes on deep learning, a literature review of methods for 6D object pose estimation from images, a description of the proposed method and achievements.