Deep Learning for the Early Detection of Diabetic Retinopathy

Abstract

One of the most prevalent eye conditions and a consequence of diabetes that affects the eyes are diabetic retinopathy. Diabetic retinopathy (DR) may simply result in minor visual issues or no symptoms at all. It may eventually result in blindness. Therefore, early symptom diagnosis may help prevent blindness. This project contributes significantly to the field of diabetic retinopathy diagnosis by providing a robust methodology that combines traditional machine learning with state-of-the-art deep learning techniques. In this work, I have provided several tests on exudate properties, blood vessel properties, and microaneurysm characteristics related to various aspects of diabetic retinopathy in two different datasets. One is gaussian filtered and another one is without any filters. Then, the stages of DR into five groups based on the features are categorized: healthy, mild dr, moderate dr, proliferative dr and severe dr. The phases are categorized using Support Vector Machine and Random

Forest classifiers and then a Convolutional Neural Network (CNN) is used to detect the five stages of diabetic retinopathy. Finally, the accuracy, sensitivity and specificity of the CNN model is calculated which are 95.28%, 93% and 94% respectively using the without any filter dataset. Thus, this initiative has the potential to both enhance patient outcomes and avoid visual loss.