Analysis and correction of OCT images for the control of robotic flexible endoscopes

This doctoral research is focused on overcoming problems in autonomous surgical procedures when instruments have to navigate towards the clinical target by accurate self-localization in the front of certain tissue and, simultaneously, to build a map of the luminal environment for medical diagnosis. Vision-based approaches using stable tissue texture are highly desirable for a wide range of applications. Optical Coherence Tomography (OCT) [1] is an imaging technique of great importance in biomedical optical applications. The backscattered light is measured of the internal structure of biological tissues to provide high resolution, axial and three-dimensional images of the sample. Endoscopic OCT catheter has been applied into cardiovascular, respiratory and digestive systems for imaging of internal structures. In gastroenterology, a balloon and capsule based catheters have been developed for imaging of the esophagus. Catheter-based imaging systems have limited Field of View (FoV), especially when considering OCT systems which emphasize more on the image resolution. For small lumens such as the vasculature and esophagus, volume reconstruction from one pull back scanning using an OCT system can be sufficient for accessing the entire lumen. However, for larger luminal environments as the colon or stomach, the link between reconstruction, robot planning and robot control needs to be established e.g. the link with robot control is needed in order to realize a certain scanning behavior, which would be necessary to make reconstruction efficient and accurate. This side-viewing catheter could be employed to actively follow the lumen wall with a robotic endoscope. The OCT augmented endoscope can provide more accurate navigation feedback for the control system. The robotic endoscope also has a camera in the distal part, which can perform a rough global navigation to aid the OCT system’s local scanning. In the local scanning process, ideally, the distance between the OCT probe and the tissue is controlled to be constant. This could keep the tissue always in the FoV of the OCT, especially for luminal tissue with a complex geometry like the colon. Another type of safe scanning mode could also be realized with contact between the OCT catheter and the colon tissue surface. In this case, a segmentation algorithm is required to provide real-time quantitative feedback about the contact or the distance. For volumetric reconstruction from the robotic scanning, computer vision and imaging processing techniques including incremental mapping or Structure from-Motion (SfM) can be deployed. The main aims could be divided into the following three: - Find an efficient configuration for robotic endoscope navigation. To achieve this task, the OCT images first need to be stabilized to improve its orientation accuracy. - Information perception for both diagnosis and navigation purpose. Tailor the machine learning based computer vision algorithm for side-viewing imaging modalities. - Design automatic scanning strategies for larger lumen environment with small FoV side-viewing probes, incorporate local navigation information with global navigation information.