Transfer Learning for Classification of Retinal Disease using Fundus Imaging

Retinal diseases, caused by both eye and systemic conditions such as diabetes and cardiovascular disease, can result in changes to the retina that can lead to blindness. Fundus imaging, which captures images of the retina, is crucial for diagnosing and treating these diseases. Utilizing Computer Aided Diagnosis (CAD) with deep learning models can help automate the detection of retinal diseases and improve care time. However, the lack of sufficient labeled data for retinal images can be a challenge. To address this, we evaluated the use of pretrained models with transfer learning for multi-label classification of retinal images. Three pre-trained models (VGG19, ResNet50, and Inception V3) are considered. The results show that multi-label classification with data augmentation and transfer learning results in improved accuracy of prediction compared to previous studies.