Recent introduction of probe-based confocal laser endomicroscopy (pCLE) enabled real-time and in-vivo visualisation of biological tissue on a microscopic level. This new technique, referred to as optical biopsy, involves the new challenge of re-targeting previous biopsy sites during surveillance examinations.
In this thesis, a novel scene recognition method is presented to support the re-targeting of optical biopsy sites in surveillance GI-endoscopies. Drawing on the mathematical framework of manifold learning, a new representation for endoscopic videos is introduced. Relying on the tools of spectral theory and their application in manifold learning, the challenge of scene recognition is approached from a novel perspective. Experimental evaluation of the methods on clinical datasets demonstrates the feasibility of the proposed methods as a promising step for assisting the endoscopic expert in re-targeting optical biopsy sites.
«
Recent introduction of probe-based confocal laser endomicroscopy (pCLE) enabled real-time and in-vivo visualisation of biological tissue on a microscopic level. This new technique, referred to as optical biopsy, involves the new challenge of re-targeting previous biopsy sites during surveillance examinations.
In this thesis, a novel scene recognition method is presented to support the re-targeting of optical biopsy sites in surveillance GI-endoscopies. Drawing on the mathematical framework of m...
»