Fully automated 3D colon segmentation for early detection of colorectal cancer based on convex formulation of the active contour model

Accurate colon segmentation would play a vital role in a virtual colonoscopy system, amounting for reliable polyp detection; a colon cancer indicator. In this paper, we propose a fully automated framework for 3D colon segmentation based on the global/convex continuous minimization of the active contour model in 3D space. For optimal results, 3D region growing and some colon anatomical features, e.g. size, persistence and curvature have been incorporated for post processing. The proposed framework, applied on 12 colon data sets, is compared with graph cuts (discrete optimization) and adaptive level sets (non-convex continuous optimization). Our results outperform the other two in different aspects including speed of convergence, sensitivity and specificity with overall accuracy of 99%.