Utilising Deep Learning Techniques, Detect and Categorise Diabetic Retinopathy

A prevalent eye disease in diabetics is the leading cause of blindness in the general population considered as diabetic retinopathy. If blood sugar levels are too high, a condition known as diabetic retinopathy develops. Diabetic retinopathy, which can worsen as diabetes progresses, may impair the patient’s vision. Subsequently, two classifications are established: proliferative diabetic retinopathy (PDR) and non-proliferative diabetic retinopathy (NPDR). Three models, including the support vector machine (SVM), are introduced, and their results were evaluated to identify diabetic retinopathy. Support vector machines kernel-based classifiers are developed and tested using a mixed database that includes recently acquired fundus imaging records from a nearby hospital as well as open-source public databases. This method employs fivefold cross-validation for the local database, and LBP and LTP features are employed to implement a fused cubic SVM classifier with 0.964% classification accuracy, 0.964% sensitivity, and 0.969% specificity. The fundus picture is grouped into distinct regions, and colour dimension is reduced by utilising a k-means colour compression algorithm. Various areas of the diabetic fundus were then segmented and analysed using the region characteristics features.