Efficient Heart Disease Risk Management: A CNN-based Approach for Automated Prediction using Retinal Fundus Image

Despite established risk factors such as diabetes, heart disease remains a major health problem. Early diagnosis of heart disease is essential for successful prevention and treatment. Traditional assessment methods can be time-consuming and complex, highlighting the need for automated detection systems. This research paper investigates using Convolutional Neural Networks (CNNs) to predict heart disease risk based on retinal fundus images automatically. We investigated different CNN designs and approaches, including developing proposed models and using pre-trained models such as EfficientNetB0 and B7. Experimental results show that a Heart Disease Prediction (HDP-CNN) model that incorporates a special feature (focal loss) and a preprocessing technique (CLAHE) to improve image clarity achieves superior performance. This method achieved a validation accuracy of 84% with remarkable efficiency. CLAHE preprocessing plays a key role in improving image clarity and enables more accurate feature extraction, which is crucial for accurate risk classification. This research highlights the potential of HDP-CNNs for automatic prediction of heart disease risk and offers a promising avenue for early detection of heart disease and efficient management.