Simultaneous Localisation and Mapping (SLAM) of The Airways in Lung Bronchoscopy

The proposed project aims to create a real-time 3D map of the lung airways, using feedback from an endoscopic camera located at the tip of a robotic bronchoscope, while simultaneously tracking the bronchoscope within the lung and localize it in the developed 3D map.



The first stage of the research will focus on the localization and shape estimation of a robotic bronchoscope. The robot consists of several nested pre-curved thin tubes, which can be manipulated by rotation and translation, ultimately changing the overall shape of the robot. The goal is to obtain a method which could be used to accurately estimate the shape of the robot in real time. The methodology is based on Kalman filtering and sensor fusion, using a combination of an existing kinematic model of the robot and additional position sensors (e.g, CT scan images).



The second stage aims to create a mapping of the bronchial tree. To achieve this, state-of-art computer vision and machine learning methods would be investigated, while making use of lung CT scans, real-life bronchoscopy video data and relevant image processing algorithms to segment the bronchial tree inside the lung. The focus would be on deep learning methods, since the traditional approaches such as feature matching have not performed well on the problems of similar nature.



Having an accurate 3D mapping of the lungs, as well as localization of the robotic bronchoscope, would be a step closer towards autonomous navigation of the bronchoscope. The potential benefits of this are improved control and reduced human error, which in turn have been associated with higher success rate of the procedures, lower costs, and shortened recovery time for the patients.