Colon fold contour estimation for 3D visualization of colon structure from 2D colonoscopy images

This paper presents a novel colon fold contour estimation technique, a crucial step toward fully-automated reconstruction of a 3D colon segment from an image captured from colonoscopy for automated labeling of unseen mucosa areas (where polyps or abnormal lesions may reside) during colonoscopy. The endoscopist can then manipulate the endoscope to inspect these unseen areas. The proposed technique estimates colon fold contours taking advantage of both computer graphics and image processing. The algorithm detects edges of the colon folds, filters noise edges, pairs edges from the same colon fold, and finally completes the contours. The completed colon fold contours and pixel properties around the contours are used to reconstruct a 3D virtual colon segment. We evaluate recall and precision of the algorithm and similarity between algorithm estimated fold contours and manually drawn fold contours.