Diabetic Retinopathy Detection using Deep Learning Framework and Explainable Artificial Intelligence Technique *

Abstract

Diabetic retinopathy (DR) is a serious eye condition induced by diabetes that can result in eyesight. In order to overcome the vision impairment of diabetes mellitus patients, it is essential for early age prevention by medical practitioners. In the traditional approach, ophthalmologists conduct various screening tests to detect DR but fail to achieve accurate and conclusive diagnosis due to the associated time consuming processes. In order to eliminate such burden on the ophthalmologists, deep learning and machine learning techniques have evolved rapidly playing a significant role in the classification of diabetic and non diabetic patients. In this paper, the proposed framework implements a publicly available Benchmark APTOS 2019 Gaussian-filtered DR image dataset using a customized CNN model, yielding an enhanced accuracy of 98 %. In addition, the framework also employs a LIME explainer visualization model to provide enhanced transparency and interpretability to the generated predictions. This eliminates the "blackbox" -ed nature of the predicted results from the traditional ML process and enables healthcare providers to take clinical decisions with confidence providing visual representations of the significant features that contribute towards an outcome.