Smart Healthcare System for Detection of Diabetic Retinopathy Using Transfer Learning

The Diabetic Retinopathy (DR) causes significant vision loss, especially among the working-age population. Manual diagnosis by Ophthalmologists is slow due to a shortage of skilled clinicians. An automated system for early DR detection is needed. Deep Learning (DL) shows promise in medical image analysis but requires ample training images, challenging in the DR domain. Transfer Learning (TL) addresses this data scarcity issue, combining TL with CNNs for DR detection. This study proposes a smart healthcare model using EfficientNetV1 and V2 CNNs with TL on IDRiD dataset for DR detection. EfficientNetV1 (B0-B7) performance is compared, evaluating sensitivity, specificity, and accuracy. EfficientNetV2's performance on the same dataset is also assessed. EfficientNetV2 being a smaller model size achieved 80.09% sensitivity, establish it as a state-of-the-art model that excels in both training efficiency and parameter efficiency compared to its predecessors, a novel model for DR detection and classification. The study showcases advancements in early DR diagnosis, promising broader accessibility. Future efforts will focus on enhancing model performance and deploying it in smart healthcare settings. The proposed smart healthcare model holds significant promise for DR diagnosis and treatment.