Optimized Technique for Detection of Diabetic Retinopathy using Segmented Retinal Blood Vessels

摘要

This paper presents an algorithm that will segment the retinal blood vessels with an accuracy of 96.17%. This algorithm will extract the features from input images present in STARE and CHASE_DB1 databases. The extracted features will be large in number, but all the features are not useful. So, the feature optimization is done by Lion Optimization which has effectively chosen only the features which are useful in representing the extracted features as blood vessels or non-blood vessels. The algorithm was applied first on training images which have results of manually segmented images already. Then the algorithm was implemented on training images and evaluated on training images and it successfully detects the normal as well as abnormal images. The quantitative results were checked using parameters sensitivity, specificity, accuracy, positive predictive rate and false predictive rate and proved to give better results in comparison to existing techniques.