Cardiovascular risk prediction using retinal images has emerged as a promising area in medical imaging. This study introduces RetinaViTNet, a novel framework that uses vision transformers (ViT) for feature extraction and classification. Unlike traditional convolutional architectures, ViTs capture long-range dependencies within retinal images, enabling the identification of subtle patterns linked to cardiovascular risks. Experimental results demonstrate that RetinaViTNet delivers superior accuracy and generalization compared to baseline convolutional models. These findings highlight the potential of RetinaViTNet to advance early cardiovascular risk assessment, offering significant promise for its application in clinical practice.