Diabetic retinopathy detection using EADBSC and improved dilated ensemble CNN-based classification

Nowadays high-rated and active research in the ophthalmic field has paved the way for the detection of several retinal disorders which helps the ophthalmologist, in scheduling and executing timely treatment. Due to more number of retinal defective patients and less number of medical professionals regular screening is one the most difficult tasks. The primary objective of this research work is to help doctors and other medical professionals predict the diabetic retinopathy. Here in this research article work retina fundus images are taken from the both Public dataset and the in-house clinical dataset from Chaithanya Eye Hospital Kerala. The first stage is to remove noise from the input image and enhance the contrast of the images. For noise reduction, a Bilateral Filter is utilized first, followed by enhancement utilizing Contrast Limited Adaptive Histogram Equalization with an unsharp technique. Then Thick Blood vessels are segmented from the enhanced image using the Extended Adaptive Density-Based Spatial Clustering (EADBSC) Method. The segmented image is fed to the classifier such as Ensemble CNN & Improved dilated ensemble CNN classification, which classifies the image as Diabetic Retinopathy (DR) or Normal case. Using Ensemble CNN, an accuracy of 97.46% is obtained. Improved dilated ensemble CNN-based classifier has an accuracy of 99.19%. By way of their efficiency estimation, the improved dilated ensemble CNN classifier is higher than most standard classifiers. It is also hoped that the developed automatic detection techniques will assist clinicians to diagnose Diabetic Retinopathy at an early stage.