Image Registration, Color Correction, and Change Detection Based on Value of Difference in Sequential Ocular Fundus Images

摘要

The purpose of this study is to assist ophthalmologic screening in health checks. A decision process for detecting progressive change is proposed in which normal cases are rejected on the basis of comparison between the ocular fundus image to be examined and the normal ocular fundus image of the same subject. Ocular fundus images contain large individual differences, but there is no progressive change in the normal condition. A large change in the image is observed only when there is a pathological change, such as white spots or hemorrhage. Consequently, an automatic screening procedure is considered in which an ocular fundus image diagnosed as normal is prepared for each individual, and images are judged to be normal when no progressive change is indicated by the difference between images. Specifically, an algorithm which consists of the following procedures is created. Image registration is applied, considering the ocular fundus shape of the individual. Color correction is applied to correct the brightness and color difference due to imaging conditions. The images are compared and the difference between the images is calculated and the presence or absence of progressive changes is decided. The algorithm was evaluated by using real ocular fundus images, and a rejection rate of 78% for normal cases was obtained. Thus, it is possible to assist screening by providing a pathological index.