Analysis of early diabetic retinopathy by computer processing of fundus images--a preliminary study

A number of classification schemes have developed for assessment of fundus changes in the early stages of diabetic retinopathy. Generally, these rely on visual inspection of fundus photographs by trained observers who estimate the degree of retinopathy compared with a set of standard photographs. Computer image processing techniques offer increased accuracy for monitoring subtle changes in retinopathy level and the possibility of more precise classification. The application of image enhancement and analysis to some discrete fundus lesions is demonstrated. Fundus photographs from the ETDRS Classification Scheme are digitized and processed to improve detectability of discrete lesions. Selected areas of an image containing haemorrhages or hard exudates are then magnified, and techniques of spatial filtering and thresholding are used to extract the features of interest from their background. The area occupied by the lesions may then be easily quantified. Results indicate that image processing methods show promise for the improved analysis of retinopathic changes.