Computer-aided Classification of Endoscopic Images from the Gastrointestinal Tract

In modern medicine endoscopy plays a very important role as it allows physicians to detect severe diseases in early development stages already. Especially the gastrointestinal tract is ex-amined routinely in order to detect pre-malignant and possibly malignant diseases. As a con-sequence, the mortality rate for many diseases, especially different types of cancers, has been lowered drastically throughout the last years. In this cumulative thesis we specifically focus on computer-aided decision support systems targeting the gastrointestinal tract as those may potentially help to save time, lower the cost of endoscopic procedures, and lower the risk of such procedures for the patients. To underpin the importance of such systems this thesis includes a detailed overview and discussion of methods published within the two past decades, targeting this field of research. Furthermore, we evaluate different types of features and classifiers throughout this work, in order to obtain a robust system for the classification of endoscopic images. The features used throughout this work either investigate local texture properties, are based on statistical texture features, or aim at a shape-based classification