Computer-aided diagnosis in virtual colonography via combination of surface normal and sphere fitting methods

The success of CT colonography (CTC) depends on appropriate tools for quick and accurate diagnostic reading. Current advancements in computer technology have the potential to bring such tools even to personal computer level. In this paper a technique for computed-aided diagnosis (CAD) using CT colonography is described. The method uses a combination of surface normal and sphere fitting methods to label positions in the volume data, which have a strong likelihood of being polyps, and presents them in a user-friendly way. The method was tested on a study group of 18 patients and the detection rate for polyps of 10 mm or larger was 100%, comparable to that of human readers. The price paid for a high detection rate was a large number of approximately eight false-positive findings per case. Our results show that CAD is feasible, and if the number of false positives is further reduced, then this method can be useful for clinical screenings.