Computer Aided Diagnosis for Virtual Colonography by Geometric Model Fitting

We present a system for Computer Aided Diagnosis in Virtual Colonography based on simple geometric model fitting. Our approach extends surface normal analysis and sphere fitting methods. We label locations in the volume data, which have a high probability of being colonic polyps, and present them in a user-friendly way. The method was tested on a study group of 30 patients. Using normal colonoscopy as standard of reference, true positive and false positive findings were determined. The detection rate for polyps larger than 9mm was 100% comparable to that of human readers. We introduce sphere fitting and normal distribution analysis as additional steps to reduce the number of false positive findings generally encountered with surface normal methods. Initial results show that Computer Aided Diagnosis is feasible and that our method holds potential for screening purposes.