Enhancing Cardiovascular Risk Prediction with Vision Transformer and Retinal Image Analysis

Cardiovascular diseases (CVD) are among the lead-ing causes of deaths. Early prediction of cardiovascular risk can play an important role in the development of improved preventive medicine. This study investigates the feasibility of using deep learning methods with retinal images for cardiovascular risk prediction. To that end, we compare the RNN and ViT models in the analysis of retinal images. Preprocessing techniques and data augmentation were also used to improve model training. The RNN and ViT models are trained and tested on various metrics, such as accuracy and ROC curves. From the results of our experiments, it can be demonstrated that the ViT model performs better compared with the RNN model in predicting the cardiovascular risk by obtaining higher accuracy and a better performance in the ROC curve. This study demonstrates the potential utility of retinal imaging in cardiovascular risk assessment, where the ViT model exhibits capabilities superior to classical RNN models. This research places a special emphasis on higher-order deep learning models, which should be used in the early stages of diagnosis and prevention in healthcare settings.