Bronco-Vision

This research project aims to explore the applicability of computer vision and machine learning techniques in intracorporeal surgical procedures, with a focus on bronchoscopy. Bronchoscopy is a minimally invasive endobronchial procedure used to examine the airways and diagnose or treat pulmonary diseases, using a thin tube with a camera and light source inserted through the mouth or nose.

The project leverages recent advances in computer vision to automate tasks such as tool recognition, anatomical structure detection, and step identification during surgery. A key goal is to create a dataset of bronchoscopy procedures, including video sequences from the bronchoscope and corresponding preoperative CT scans. The CT scans will be segmented to produce accurate 3D reconstructions of the bronchial tree, serving as ground truth to validate computer-generated 3D models.

Semantic segmentation algorithms will be developed to identify relevant anatomical regions and bifurcations in the bronchial tree, enabling precise localization of the bronchoscope. The dataset will be used to quantitatively evaluate state-of-the-art vision techniques in this context.

Scientifically, the project aims to provide valuable data for evaluating computer vision in internal surgeries. Medically, it could reduce procedure time and dependency on CT scans by enabling real-time 3D modeling. Socially and economically, it supports the advancement of personalized and robotic medicine.