GPU-accelerated Polyp Detection in Virtual Colonoscopy

Nowadays, Virtual Colonoscopy (VC) is an important non-invasive alternative for the study of the colon. Substantial research efforts have been dedicated to this method, and one of the major challenges has always been producing accurate results in a short period of time. One of the most crucial phases of VC is the detection of polyp candidates, where possible lesions on the colon walls are automatically detected. Frequently, this stage requires intensive computations and therefore it is important to develop new techniques for reducing its execution time. This paper presents a technique for automatic detection of polyp candidates based on curvature analysis that reduces the execution time using parallel programming in CUDA. Additionally, we introduce a novel technique for discarding false positive detections based on the shape of a candidate in a planar cut. The obtained results show a remarkable reduction in execution time with respect to a CPU implementation as well as a low rate of false positives.