Abstract: To explore the feasibility of computer aided diagnosis in virtual colonoscopy by means of 3D texture analysis, 190 lesion data volumes are manually extracted and 39 texture features are computed from each lesion to form a feature vector. All polyps are grouped into four pathological types of hyperplastic (H), tubular adenoma (Ta), tubulovillous adenoma (Va), and adenocarcinoma (A). The Hotelling T-square test on all the feature vectors shows that significant differences are found between all the paired groups except the H/Ta group. This result indicates the possibility of computer aided diagnosis in virtual colonoscopy when the proposed 3D textural features are used in the classification of four different types of lesions.

Key words: virtual colonoscopy, computer aided diagnosis, three dimension texture analysis