Colon cancer is the second leading cause of cancer related deaths in the developed nations. Early detection and removal of colorectal polyps via screening is the most effective way to reduce colorectal cancer (CRC) mortality. Computed Tomography Colonography (CTC) or Virtual Colonoscopy (VC) is a rapidly evolving non-invasive technique and the medical community view this medical procedure as an alternative to the standard colonoscopy for the detection of colonic polyps. In CTC the first step for automatic polyp detection for 3D visualization of the colon structure and automatic polyp detection addresses the segmentation of the colon lumen. The segmentation of colon lumen is far from a trivial task as in practice many datasets are collapsed due to incorrect patient preparation or blockages caused by residual water/materials left in the colon. In this thesis a robust multi-stage technique for automatic segmentation of the colon is proposed tha t maximally uses the anatomical model of a generic colon. In this regard, the colon is reconstructed using volume by length analysis, orientation, length, end points, geometrical position in the volumetric data, and gradient of the centreline of each candidate air region detected in the CT data. The proposed method was validated using a total of 151 standard dose (lOOmAs) and 13 low-dose (13mAs-40mAs) datasets and the collapsed colon surface detection was always higher than 95% with an average of 1.58% extra colonic surface inclusion. The second major step of automated CTC attempts the identification of colorectal polyps. In this thesis a robust method for polyp detection based on surface curvature analysis has been developed and evaluated. The convexity of the segmented colon surface is sampled using the surface normal intersection, Hough transform, 3D histogram, Gaussian distribution, convexity constraint and 3D region growing. For each polyp candidate surface the morphological and statistical features are extracted and the candidate surface is classified as a polyp/fold structure using a Feature Normalized Nearest Neighbourhood classifier. The devised polyp detection scheme entails a low computational overhead (typically takes 3.60 minute per dataset) and shows 100% sensitivity for polyps larger than 10mm, 92% sensitivity for polyps in the range 5 to 10mm and 64.28% sensitivity for polyp smaller than 5mm. The developed technique returns in average 4.01 false positives per dataset. The patient exposure to ionising radiation is the major concern in using CTC as a mass screening technique for colonic polyp detection. A reduction of the radiation dose will increase the level of noise during the acquisition process and as a result the quality of the CT d a ta is degraded. To fully investigate the effect of the low-dose radiation on the performance of automated polyp detection, a phantom has been developed and scanned using different radiation doses. The phantom polyps have realistic shapes (sessile, pedunculated, and flat) and sizes (3 to 20mm) and were designed to closely approximate the real polyps encountered in clinical CT data. Automatic polyp detection shows 100% sensitivity for polyps larger than 10mm and shows 95% sensitivity for polyps in the range 5 to 10mm. The developed method was applied to CT data acquired at radiation doses between 13 to 40mAs and the experimental results indicate th a t robust polyp detection can be obtained even at radiation doses as low as 13mAs.