Prognostication of diseases is extremely important because the results of the prognostic studies can aid the doctors in better decision making regarding the treatment of the patient and serve well to reduce any anxiety in the patients regarding the course of their treatment. Diabetic retinopathy is basically a complication in diabetics that is brought about by the damage to the of the eye’s blood vessels, that even though initially only causes mild problems, can eventually lead to irreversible loss of vision, and hence the early prognostication of Diabetic Retinopathy is exceedingly essential, as the early detection of this disease can reduce the risk or even avoid irreversible loss of vision. DR can be prognosticated by ophthalmologists with the help of color fundus images but that can cause the patient to lose a lot of valuable time. The method proposed by us in this paper is developed on CNN under deep learning that uses medical images in an unsupervised manner and aims at learning high- and low-level features which can aid in the detection, classification, feature extraction and eventually prognostication of Diabetic Retinopathy. The input images are initially pre-processed and prepared for the model by filtering, compressing and resizing the images. In this paper we successfully classify DR into Mild, Moderate, Severe Non-proliferative DR and Proliferative DR. The advantage of this method over others is the increased efficiency over other methods. It outperforms other methods if tested under the same condition, and it also trains faster and performs better. In the future we plan to use alternate neural network methods such as Probabilistic Neural Network and increase the estimations.

TLBO, Diabetic Retinopathy, CNN, Machine Learning, SVM