Meta methods for computer-aided endoscopic image analysis

Video endoscopy is an invaluable diagnostic tool for disease prediction in internal cavities and hollow organs such as the gastrointestinal tract, the respiratory tract, the ear, the urinary tract and the female reproductive system. This diagnostic procedure involves the introduction of an imaging device into the body to capture images of the internal structures. The clinician examines these images and mentally performs tasks such as image matching and characterization to determine a diagnosis. Advances in computer vision and machine learning can be applied to create tools for computer-aided diagnosis which can help improve clinical care by reducing the diagnostic time or providing relevant feedback to assist the clinician. However, direct application of these techniques to endoscopic imagery is not a trivial task due to the complexity of endoscopic scenes and limitations of the imaging systems.In this dissertation, we describe steps towards computer-aided diagnosis of endoscopic images. We hypothesize that the challenges presented by endoscopic imagery can be addressed by meta systems, which integrate information from multiple sources. We propose such meta methods for image matching, classification and mosaicking and demonstrate their applicability to augmenting Crohn's Disease evaluation with wireless capsule endoscopy and endometrial cancer diagnosis with contact hysteroscopy. Validation results on simulated, phantom and real patient data are presented.