Protocol and support infrastructure for the creation of an annotated database of images by confocal endomicroscopy

• Autor(a) principal: Brito, Aniana da Rosa de
• Data de Publicação: 2011
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10316/18327
• topic: Microscopia - confocal; Endomiscroscopia aquisição de imagem CEM; Gastroenterologia

Confocal laser endomicroscopy (CEM) is a new diagnosis technique that enables the histological examination of suspicious tissues in real-time during an ongoing endoscopy. The main advantage of the technique is that it avoids tissue biopsy for lab analysis, providing the doctor with the means to make an immediate diagnosis. Two main difficulties can be found in endomicroscopy exam: (i) the simultaneous execution of endoscopy and histology exam is very difficult to accomplish in practice and requiring the doctor to go through a very long training period; (ii) this technique only recently was adapted for in vivo histological examination, so the image taxonomy and interpretation is not well defined yet. The overall goal of this line of research is to develop a computational system that applies pattern recognition techniques for assisting the practitioner during the procedure by performing automatic diagnosis from the CEM images. The main goal of this study is to develop software for collecting and labeling CEM images to build a database, and use this data to develop basic algorithms for detection and segmentation of the cellular structures. In this thesis we describe the image acquisition protocols and the application developed to acquire the data obtained by endomicroscopy. Regarding the data analysis, a statistical and semantic analysis of images was performed. The statistical analyses of database show that some characteristics like the number and shape of the crypts, allow distinguishing the different classes of the image. The results obtained from semantic analyses, done through texture analyses, show that it does not allow distinguishing the image classes. As demonstrated the statistical analyses the number and shape of the crypts are the parameters that enable to distinguish the classes. Therefore in this project we segment and detect the crypts. Relatively to the crypts segmentation results, we concluded that the symmetry energy work well in image of normal tissue and image with light inflammation. So it is needed to take into consideration much more information and features. Keywords: Confocal microscopy, endomicroscopy, application, database, segmentation.