Cardiovascular Disease Prediction from Retinal Images Using Mobilenet and Deep Learning

Abstract

CVDs, in fact, are still among the leading causes of death in the world. Emerging scientific evidence suggests that retinal abnormalities may be used as biomarkers for systemic diseases like CVDs. Herein, we propose a deep learning methodology for diagnosis using CNN with MobileNet as a backbone to classify retinal images into six categories: Age-Related Macular Degeneration (ARMD), Diabetic Neuropathy (DN), Diabetic Retinopathy (DR), Macular Hole (MH), Optic Disc Cupping (ODC), and Normal. These conditions carry implications for the systemic health of the patient, somewhat like CVDs. The network was trained with large heterogeneous retinal fundus image datasets, which included preprocessing steps such as resizing, normalization, and data augmentation. Owing to MobileNet's lightweight architecture, it has a good potential for easy and scalable deployments, particularly in resource-limited clinical environments. All model performances and evaluations are assessed using accuracy, precision, recall, F1 score, and confusion matrix metrics. Results show a powerful potential for the model to separate different retinal diseases from the normal state, for risk calculation, and for the purposes of early detection strategies. Future aspects will include clinical validation and the integration of the findings into diagnostic pathways to allow full-scale non-invasive patient screening with the assistance of health associates.