3D Shape Reconstruction from Capsule Endoscopy Video
Abstract
Wireless Capsule Endoscopy (WCE) has revolutionized the field of gastroenterology by providing an alternative to traditional endoscopy, as it is less invasive and requires no sedation. It is a patient-friendly procedure to visualize the entire gastrointestinal (GI) tract, from the esophagus to the large intestine, using a small, swallowable capsule equipped with a miniature camera. Despite its many advantages, WCE images also entail several challenges which include issues related to uneven and low illumination, low resolution, and noise. Moreover, the lack of control over the capsule's movement within the GI tract restricts the thorough examination of areas of particular interest.
This thesis endeavors to address some of the challenges in WCE domain by generating 3D models of GI tract, enhancement of WCE images and emphasizing the practical utility of these models and enhanced images in clinical evaluation. Through a systematic exploration of 3D reconstruction methodologies, image enhancement strategies, and clinical applications, this thesis achieves its three primary objectives. Notably, the evaluation of the Shape from Shading (SFS) method provides insights into its applicability to current WCE technologies. The study encompasses the implementation of 3D reconstruction technique on a spectrum of scenarios, ranging from controlled environments to synthetic colon imaging, and culminating in the application to endoscopic and WCE images, achieving the foundational objectives. Anticipating future advancements in WCE technology, the research proposes innovative methodologies, particularly highlighting the effectiveness of the Shape from Focus (SFF) in GI shape recovery. Addressing issues of uneven illumination, a lighting conversion strategy enhances image quality for diagnostic purposes, with successful application to real endoscopic images. The research concludes by emphasizing the foundational role of these findings in shaping the future of clinical decision support systems in WCE, validating the potential of 3D models and enhanced images as valuable diagnostic tools in gastroenterology.
The methodologies introduced in this thesis have the potential to influence not only the field of WCE, but also extend to other domains. These proposed techniques aim to tackle key challenges in 3D reconstruction and image enhancement. The outcomes could significantly benefit gastroenterologists by offering improved visualizations, thereby enhancing diagnostic accuracy and patient care.