Nawaz, Slabaugh: Global Analysis of Haustral Fold Ridges

Haustral folds are a major source of false positives (FPs) in the Computer Aided Detection (CAD) of colorectal lesions. In this paper, a novel bottom-up data-driven approach is presented to globally analyze haustral folds. Unlike the earlier haustral fold detection methods that rely solely on local differential geometric computations, which can be unreliable, particularly for thicker folds, our method builds a global model of the fold ridge to characterize the fold on a much larger scale. We propose a novel geometric feature, the density of ridge candidates (DRC), to remove CAD false positives. The method is tested on a dataset of 333 CAD marks detected using a state-of-the-art CAD system [7] on a data from 34 patients, and a significant FP reduction of 14.13% is achieved while maintaining 100% sensitivity.