Performance Evaluation of AI Assisted Automotive Diabetic Retinopathy Classification Systems

Abstract

The reliable diagnosis of diabetic retinopathy (DR) has long been a source of concern for researchers. Due to fluctuating glucose levels, the blood vessels in the retina are more vulnerable to aberrant metabolism. These variances result in lesions or retinal damage, which are then referred to as DR collectively. The signs of DR are often difficult for the current eye healthcare procedures to diagnose. Building an artificial intelligence-assisted automated DR classification (AI-ADRC) system is an excellent way to reduce the pressure of incorrect diagnoses as a result. This article is focused on performance evaluation of DR classification methods, which includes machine learning models, deep learning models, feature extraction, and feature selection methods. The problems presented in state-of-art AI-ADRC systems are addressed, which will help to develop the novel AI-ADRC model. Further, the deep learning-based AI-ADRC models are resulted in superior performance as compared to machine learning based AI-ADRC models using various datasets.