Predicting Heart Attack Risk From Retinal Fundus Images by Using the EfficientNet-B0 Model

Cardiovascular diseases remain a leading cause of mortality worldwide, with early detection being critical for effective intervention. The non-invasive nature of retinal fundus imaging allows doctors to identify microvascular alterations, which serve as indicators for assessing the risk of developing coronary heart disease. In this study, a novel framework was proposed for predicting the risk of heart attack using retinal fundus images combined with EfficientNet-B0, a deep convolutional neural network architecture optimized for both accuracy and computational efficiency. Fundus images were collected and preprocessed through resizing, normalization, and data augmentation techniques to enhance model generalization. Key vascular and structural features were extracted and transformed into feature embeddings, which were then fed into the EfficientNet-B0 model to generate hierarchical feature maps. Fully connected layers with ReLU activation and a final SoftMax classifier were employed to predict high- and low-risk cardiac conditions. The model was trained and validated using annotated retinal datasets from clinical studies spanning multiple years, with cross-fold validation applied to ensure robustness. The experimental results demonstrate that the proposed framework achieved high predictive performance, capturing subtle vascular abnormalities correlated with cardiovascular risk. The system achieves an accuracy rate of 98.92%, indicating that it can serve as a non-invasive, automated screening system for detecting cardiovascular problems in their early stages. The results demonstrate that combining retinal imaging with deep learning technology produces effective methods for early cardiac health monitoring.