Automated Texture-Based Identification of Ovarian Cancer in Confocal Microendoscope Images

摘要

The fluorescence confocal microendoscope provides high-resolution, in-vivo imaging of cellular pathology duringoptical biopsy. There are indications that the examination of human ovaries with this instrument has diagnosticimplications for the early detection of ovarian cancer. The purpose of this study was to develop a computer-aidedsystem to facilitate the identification of ovarian cancer from digital images captured with the confocal microendoscopesystem. To achieve this goal, we modeled the cellular-level structure present in these images as texture and extractedfeatures based on first-order statistics, spatial gray-level dependence matrices, and spatial-frequency content. Selectionof the best features for classification was performed using traditional feature selection techniques including stepwisediscriminant analysis, forward sequential search, a non-parametric method, principal component analysis, and aheuristic technique that combines the results of these methods. The best set of features selected was used forclassification, and performance of various machine classifiers was compared by analyzing the areas under their receiveroperating characteristic curves. The results show that it is possible to automatically identify patients with ovarian cancerbased on texture features extracted from confocal microendoscope images and that the machine performance is superiorto that of the human observer.