Automated colonic polyp detection using computed tomography data.

摘要

This thesis presents a novel approach to the detection colonic polyps using Computed Tomography (CT) data. Computer-aided detection of colonic polyps can be used to improve the efficiency and accuracy polyp screening. Automated polyp detection is a difficult problem due to the convoluted nature of the colon and the many structures found in the colon that can mimic polyp shape, such as haustral folds, stool, and the ileocecal valve. Haustral folds constitute the greatest challenge due to their varying shapes, sizes and prevalence throughout the colon leading to a large number of false positives. In addition, polyps lying along the side or on top of folds tend to be masked.; In this thesis we model the surface of the colon using a local geodesic polar coordinate system based on the exponential map. To simply the continuous map, a piece-wise linear model is constructed and parameters describing the local surface shape are extracted. These parameters are independent of location and orientation within the colon. Through further analysis of these parameters, six features are generated to help distinguish polyp surfaces from non-polyp surfaces. We evaluate the performance of the system, using two performance measures, a segmentation performance measure and a detection performance measure. The first measures how well the classifier signals the existence of a polyp; and the second measures how much of the polyp has been correctly classified.; Experimental data sets consisted of 9 colons containing a total of 20 polyps, ranging in size from 4mm to 30mm. The location and size of each polyp was determined by an expert radiologist using the axial CT slices. A binary decision tree classifier is used to classify the surfaces as polyp or non polyp. To reduce the number of false positive classifications the tree is pruned using a validation set, thereby reducing its complexity. Further reduction of false positives is done by grouping "polyp" classifications by their location in the colon. This shows that a sensitive and specific method of CAD for colonic polyps is possible.