CONCAVITY ANALYSIS FOR REDUCTION OF ILEOCECAL VALVE FALSE POSITIVES IN CTC

The ileocecal valve (ICV) is a common source of false-positive (FP) detections in CT colonography (CTC) computer-aided detection (CAD) of polyps. In this paper, we propose an automatic method to identify ICV CAD regions to reduce FPs. The ICV is a particularly challenging structure to detect due to its variable, polyp-mimicking morphology. However, the vast majority of ICVs have a visible orifice, which appears as a 3D concave region. Our method identifies the orifice concave region using a partial differential equation (PDE) based on 3D curvature and geometric constraints. These orifice features, combined with intensity and shape features generated in a Bayesian framework, comprise a set of compact features fed into an Adaboost classifier to produce a final classification of a region being ICV or non-ICV. Experimental results on a multi-center tagged CTC dataset demonstrate the success of the method in detecting ICV regions and reducing FPs in CAD.