Curvelet Transform and ISODATA Thresholding for Retinal Vessel Extraction

Researchers says eyes are good indicator of many diseases like Diabetic retinopathy, Glaucoma, Hyper tension, cardiac disease and many age related abnormalities. Here we aim to put forward a reliable, fast and computerized method to get the vessel network of fundus image, that can assist the ophthalmologist to diagnose the disease in early stage and also help in keeping the track of progress of treatment. Here curvelet transform is applied to the green channel of retinal image. It highlights the curves and edges of the vessel. Then CLAHE is implemented to improve the contrast and take care of the background inhomogeneity. Then the vessels are extracted by a simple ISODATA thresholding. This method gives 95.12% accuracy and average sensitivity, precision and specificity values 72.14%,73.52% and 97.35% respectively. The execution time is also very low i.e. 1.36 s. This approach is very fast and shows a very high value of accuracy while retaining comparable values of sensitivity specificity and precision.