Computer-aided detection of polyps in CT colonography

Abstract

We have developed a computer-aided diagnosis system for the detection of colonic polyps. Our system allows radiologists to detect polyps using a virtual flythrough within the 3D model of the reconstructed colon. More importantly, it includes an automatic polyp detection scheme that automatically highlights suspicious regions. As a first reader, this polyp detection scheme reduces interpretation time and decreases inter-observer variability among different radiologists. Polyp candidates are first identified using shape index and curvedness, subsequently filtered by a simple set of rules, before being subjected to linear discriminant analysis (LDA). Optimal subset of features was selected by means of a genetic algorithm. 45 fecal-tagged scans which include 71 polyps (at least 5mm in diameter) were obtained; a 5-fold cross-validation was used to estimate the generalizability of our polyp detection scheme. With sensitivities at 60% and 90%, the respective average numbers of false positives per scan are 6.22 and 18.94.