A Model-Driven Bayesian Method for Polyp Detection and False Positive Suppression in CT Colonography Computer-Aided Detection

Abstract

This chapter presents an automated method to identify colonic polyps and suppress false positives for Computer-Aided Detection (CAD) in CT Colonography (CTC). The method formulates the problem of polyp detection as a probability calculation through a unified Bayesian statistical approach. The polyp likelihood is modeled with a combination of shape, intensity, and location features, while also taking into account the spatial prior probability encoded by a Markov Random Field. A second principal curvature PDE provides a shape model; and partial volume effect is incorporated in the intensity model. When evaluated on a large multi-center dataset of colonic CT scans, the CAD detection performance as well as the volume overlap ratio demonstrate the potential of the proposed method. The method results in an average 24% reduction of false positives with no impact on sensitivity. The method is also applicable to generation of initial candidates for CTC CAD with high detection sensitivity and relatively lower false positives, compared to other non-Bayesian methods.