Bran AI: An Integrated Approach to Heart Attack Risk Prediction Using Retinal Fundus Image

This study introduces Bran AI, an advanced framework for heart attack prediction, using retinal fundus images integrated with clinical data to assess cardiovascular risk. The system's core employs a neural network-based segmentation model for detailed image analysis, enabling the extraction of critical biomarkers. Bran AI combines this with BERT for medical metadata processing, capturing relevant factors such as age, sex, and diabetes status to enhance prediction accuracy. The model architecture integrates dense layers and attention mechanisms to optimize prediction, while explainability tools such as SHAP and Grad-CAM clarify the model's decision-making process for clinical applicability. Recent studies have demonstrated the efficacy of retinal image analysis in cardiovascular risk prediction, underscoring its potential as a diagnostic tool. Bran AI achieves a high predictive performance with 92.4% accuracy, demonstrated by AUC-ROC and F1-score metrics. This innovative approach positions Bran AI as a valuable tool for healthcare professionals in cardiovascular diagnostics, fostering early detection and personalized management of heart attack risks.