Automated Detection of Diabetic Retinopathy: A Comparative Study of Machine Learning Algorithms

Abstract

Diabetic Retinopathy (DR) is a disorder of the eye and refers to the damage to the blood vessels of the retina as a result of high blood sugar levels in the body. This condition is the most common cause of blindness among working-age people. Vision impairment may result from DR and is regarded as a serious diabetes complication all over the globe. This paper evaluates the efficacy of two deep learning models, DenseNet-121 and ResNet-50, which have a widespread application in performing automated analysis of retinal images and detecting the presence of DR. DenseNet utilizes dense connectivity in order to efficiently reuse features, while ResNet uses residual connections to enhance the training of deep networks. The experiments were conducted on both models using an open-sourced DR dataset and their performance was evaluated with respect to accuracy, sensitivity, specificity and computational efficiency. The results of the analysis suggest that DenseNet is superior to ResNet in terms of accuracy and parameter efficiency, and therefore it is the best method in dealing with DR in a clinical setting. This information may assist the physicians in determining the appropriate models which should be employed for diabetic retinopathy detection in clinics.