The importance of image quality assessment for the creation of reference image collection for computer-assisted diagnosis in colonoscopy

Many applications in the field of computer assisted diagnosis (CADx) and detection (CADe) are based on reference image col- lections. These collections contain images depicting e.g. benign and malign tissues to be used for content-based image retrieval systems in order to classify the contents of yet unknown (query) images. The distinction is usually based on features calculated from these images, where typically textural features from first and second order statistics are applied. In order to obtain the best results with respect to the classification task, we believe that the related quality of the images is therefore a fundamental aspect to be considered. One well-known und understood aspect of image quality relates to image sharpness. Especially in endoscopic applications, the images used for reference image collections are often extracted from video streams recorded during the examination procedure. Therefore, changes in image quality between successive frames caused by the movements of the endoscope are most likely. Hence, in this paper we address the question, if approaches to estimate the image sharpness to preselect good images may have an influence for CAD application.