Towards Vessel Characterisation in the Vicinity of the Optic Disc in Digital Retinal Images

Automated image processing has the potential to assist in the early detection of diabetes, by detecting changes in blood vessel patterns in the retina. This paper describes progress towards the development of an integrated automated analyser of the retinal blood vessels in the vicinity of the optic disc using digital colour retinal images. First the optic disc was detected using a combination of Butterworth filtering, canny edge detection and morphological filters. After finding initial points using a median filter, blood vessels were tracked at one optic disc radius from the optic disc boundary, by two-dimensional fitting of a physically inspired model to a local region of a vessel. The last step involved the classification of the segmented vessels into arteries and veins, by using colour and hue features as inputs to a variety of classifier algorithms. The optic disc was located to within 2.5% of one optic-disc diameter in 13 of 20 images when compared to the manually identified optic disc centre. Using the median filter we obtained good accuracy for locating blood vessels. Optimisation of the blood vessel classifier using the naïve Bayes rule, resulted in a mean accuracy of 70% (s.d.=17.6%) over eight images analysed.