Automated classification of colon polyps in narrow band imaging colonoscopy

Mehr zum Buch

Early detection and treatment of colon cancer can significantly reduce its incidence and effects while improving 5-year survival rates. Narrow-Band Imaging (NBI) enhances the contrast between blood vessel structures of colon polyps and benign tissue, allowing medical practitioners to differentiate between benign hyperplasias and potentially malignant adenomas. This book discusses a computer-based polyp classification system that analyzes the blood vessel network to understand polyp evolution. The process begins with polyp localization, exploring various methods to automate this step, replacing manual selection. The segmentation of the blood vessel system produces a binary vessel map, from which innovative features are extracted from both the vessel network and original NBI images. These features were developed in collaboration with medical staff at University Hospital Aachen to mimic the decision-making process of practitioners. However, using all features in the Support Vector Machine classification resulted in suboptimal outcomes due to high dimensionality and redundancy. Consequently, the book investigates several feature selection strategies and conducts extensive tests to identify the most effective combination of features tailored for polyp classification.