Retinal vessel measurement using model fitting approach

Retinal vessels provide essential information for diagnosing retinal diseases and other diseases. Vessel detection and measurement in color retinal images are investigated in this paper. The source of blood vessels is identified by the matched Gaussian filter. A vessel is tracked from its source by the combined Gaussian and Kalman filters. A modified Gaussian model is applied to describe the vessel profile. The normal profiles along the vessel track are fitted to the proposed Gaussian model. The relationship between the vessel diameter and the obtained model parameter is studied for arteries and veins separately. Thirty-five retinal images were tested by the proposed algorithms. The mean difference of vessel measurement between our automatic algorithm and human grader is 7.5 μm (1.4 pixels on the image).