Abstract
Diagnosis using funduscopy images is a well-established means of early discovery of diabetic retinopathy and other eye disorders. However, issues remain in objectively and accurately ascertaining subtle variations in observations and accurately assessing their correspondence to patient vision. Accordingly, in this study, a diagnostic assistance method was developed. In this method, by using subtraction processing, differential images showing temporal differences between funduscopy images are presented. In addition, degree of temporal change as quantitatively calculated from temporal change magnitudes. Especially, in the case of funduscopy in which the forehead and chin are held motionless and the same imaging device is used, geometric changes between multiple images are limited to linear changes. Therefore, high-precision image registration is easily achieved, yielding differential images containing minimal artifacts due to mis-registration. The present method is capable of computing temporal differential images and degrees of temporal changes for any two images. In this study, the proposed method was applied to a single case (a set of twelve time-series images), in which we set initial examination's image as a reference image. The results verified that pathological changes in each differential image were enhanced. Moreover, the results also clearly indicated a high correlation between the degree of temporal changes and the patient vision at each time instance (0.912 correlation coefficient). In order to generalize this conclusion, it will be necessary to verify the effectiveness of this method by increasing the number of test cases.
