Automated Diabetic Retinopathy Detection Using Transfer Learning Models
Silvia Sanjana, Nazmus Shakib Shadin, Mayisha Farzana
Abstract
Diabetic retinopathy (DR) is one of the most leading symptoms of vision-loss globally. Early detection and screening can halt its progression. Until date, ophthalmologists have manually screened DR, however, DR detection can be difficult in low-resource areas when there are few ophthalmologists available. Deep learning has recently been one of the most popular strategies for improving performance in a variety of fields, particularly medical image analysis and classification. It can be used to more effectively detect DR and so maintain vision. Transfer learning models are becoming increasingly commonly employed as a deep learning method, and they are quite effective. Two public datasets which contain 1115 retinal fundus images are used in this research. Our research proposed a binary classification of DR, which is done with five Transfer learning models Xception, InceptionResNetV2, MobileNetV2, DenseNet121, and NASNetMobile which achieved the highest validation accuracy of 86.25%, 96.25%, 93.75%, 81.25%, and 80.00% respectively.