期刊名称：International Journal of Image Processing (IJIP)
电子版ISSN：1985-2304
出版年度：2013
卷号：7
期号：4
页码：418-429
出版社：Computer Science Journals
摘要：Flexible cystoscopy is an examination that allows physicians to look inside the bladder. In flexible cystoscopy, beginner physicians tend to lose track of the observation due to complex handling patterns of a flexible cystoscope and poor characteristics of the bladder. In this paper, as a diagnostic support tool for beginner physicians in flexible cystoscopy, we propose a system for tracking the observation using cystoscopic images. Our system discriminates three handling patterns of a flexible cystoscope, namely bending, rotation, or insertion. To discriminate the handling patterns accurately, we propose to use the degree of bending, rotation, or insertion as features for the discrimination as well as ZNCC-based optical flows. These features are learned by a Random Forest classifier. The classifier discriminates sequential handling patterns of the cystoscope by a time-series analysis. Experimental results on ten videos obtained in flexible cystoscopy show that each of the three handling patterns were correctly discriminated over 90% in average. In addition, we reproduced the observation in a virtual bladder we propose.
