Early detection and treatment of colon cancer can significantly reduce the disease's incidence and effects as well as increase 5-year-survival rates of patients. Narrow-Band Imaging (NBI) is an illumination technique used to increase the contrast between blood vessel structures of colon polyps and of benign tissue. Vessel patterns are used by medical practitioners to distinguish between benign hyperplasias and potentially malignant adenomas. The computer-based polyp classification system described in this book analyses the blood vessel network to gain information on the evolution of the polyp.
The first step is polyp localization where several different approaches are investigated to replace the manual selection by a partially or fully automatic system. Segmentation of the blood vessel system results in a binary vessel map. Subsequently, innovative features are extracted from the blood vessel network and the original NBI image data. These features were developed in close cooperation with the medical staff at University Hospital Aachen, aiming to emulate the decision process of the medical practitioners.
Using all features in the subsequent Support Vector Machine classification leads to inferior results as the number of dimensions is high and redundant information used. Therefore, several feature selection strategies are investigated and extensive tests performed to determine the best combination of the features specifically developed with polyp classification in mind.