Machine Learning-based Diabetic Retinopathy Early Detection and Classification Systems- A Survey

• Published in: 2021 1st Babylon International Conference on Information Technology and Science (BICITS) pp. 16 - 21
• Main Authors: Hasan, Dathar A., Zeebaree, Subhi R. M., Sadeeq, Mohammed A. M., Shukur, Hanan M., Zebari, Rizgar R., Alkhayyat, Ahmed H.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 28.04.2021
• Subjects: CAD; Classification; Classification algorithms; Clustering; Diabetes; Diabetic Retinopathy; Feature extraction; Machine Learning; Machine learning algorithms; Prediction algorithms; Regression; Retina; Retinopathy
• DOI: 10.1109/BICITS51482.2021.9509920

Diabetes Mellitus is a chronic disease that spreads quickly worldwide. It results from increasing the blood glucose level and causes complications in the heart, kidney, and eyes. Diabetic Retinopathy (DR) is an eye disease that refers to the bursting of blood vessels in the retina as Diabetes exacerbates. It is considered the main reason for blindness because it appears without showing any symptoms in the primitive stages. Earlier detection and classification of DR cases is a crucial step toward providing the necessary medical treatment. Recently, machine learning plays an efficient role in medical applications and computer-aided diagnosis due to the accelerated development in its algorithms. In this paper, we aim to study the performance of various machine learning algorithms-based DR detection and classification systems. These systems are trained and tested using massive amounts of retina fundus and thermal images from various publicly available datasets. These systems proved their success in tracking down the warning signs and identifying the DR severity level. The reviewed systems' results indicate that ResNet50 deep convolutional neural network was the most effective algorithm for performance metrics. The Resnet50 contains a set of feature extraction kernels that can analyze retina images to extract wealth information. We conclude that machine learning algorithms can support the physician in adopting appropriate diagnoses and treating DR cases.