Colorectal polyp segmentation using front propagation on surfaces guided by shape

Polyp size is a biomarker of colon cancer. Manual size measurements are subject to a variety of sources of error. We present an automatic method for segmenting a polyp from a user clicked point. The method is based on front propagation on surface mesh, guided by features that characterize the local protrudedness, its thickness, its resemblance to wall like structures and ridge measures. These measures are designed to characterize growths in the colonic lumen and differentiate polyp growth from other protrusions such as haustral folds. These measures are aggregated and smoothed. Fast marching guided by these features extracts the polyps. Empirical observation suggests that the method successfully segments a variety of polyp shapes in less than 2s.