Automated Analysis of Retinal Images

A digital image processing system for ophthalmic applications will be developed. The system will enable users to quantify easily, quickly, and accurately various retinal lesions from retinal images. Lesions such as macular drusen exudates, cotton wool spot, hemorrhages, and aneurysms will be detected in a fully automated manner; thus, the detection of such problems will eliminate the need for operator decision inherent in previously developed systems. The detection and segmentation of lesions will be accomplished through sophisticated application of Mathematical Morphology directly to gray scale images. This methodology enables the enforcement of both brightness and size (of the lesions) constraints in the detection process. The approach's feasibility will be investigated in Phase I by developing the detection/segmentation techniques needed to analyze drusen and by comparing the reproducibility achieved to the reproducibility obtained with previous semi-automatic and manual grading methods. In Phase II of this project the techniques will be extended to many other lesions. In addition, we will develop, implement and test techniques to characterize the distribution of lesions and their spatial organization. Methods to register retinal images taken at different times will be developed and used to characterize the changes in lesions over time.