Automatic Detection of Microaneurysms Using Modified UNET Architecture

Abstract

Increasingly, more people throughout the world are suffering from Diabetic Retinopathy (DR). A high blood glucose level damages the retina, which is found at the back of the eye and causes vision loss, leading to DR. The earliest symptoms of DR are referred to as microaneurysms (MAs). Due to their small size, dusky color, and practically round shape, these MAs are easy for ophthalmologists to overlook during physical investigation. In this situation, reliable early MA diagnosis is useful to prevent DR before irreversible blindness. The manual detection of DR is a labor-intensive and time-consuming process. This manual detection leads to misclassification of DR images often. The proposed method uses modified UNET architecture to detect MAs in the IDRiD Dataset and yields an accuracy of 99.88%, a precision of 0.64768, an F1 score of 0.6765, and an AUC of 0.6765.