Smart Phone-Based Fundus Imaging and Transformer-Integrated Machine Learning for Enhanced Retinal Health Care Access

Retinal diseases such as glaucoma and diabetic retinopathy often go undiagnosed due to barriers in traditional screening methods, resulting in preventable vision loss. To address this issue, we propose a smart phone-based solution that includes an add-on device for capturing fundus images, integrated with the AEResNet50 model, which presents a novel approach by seamlessly combining the strengths of ResNet50 and transformer techniques. This innovative model classifies retinal images into four categories: glaucoma, diabetic retinopathy, cataract, and normal. By enabling self-administered tests, our approach significantly enhances access to retinal health care and facilitates early disease detection while outperforming previous models in accuracy. Future work will focus on integrating this model with smart phone imaging systems for real-world deployment, advancing global health initiatives and supporting the UN's Sustainable Development Goals.