Heart Disease Detection using Vision-Based Transformer Ensemble Models

This paper addresses the challenge of accurately detecting heart disease from electrocardiogram images using a vision-based machine-learning approach. We propose a novel ensemble model that combines Vision Transformers with Residual Neural Network architectures to enhance detection accuracy across multiple heart conditions. Using a 12-lead electrocardiogram image dataset encompassing five distinct heart condition categories, our approach involves comprehensive image preprocessing and implementing two ensemble models: Vision Transformers + Residual Neural Network 18 and Vision Transformers + Residual Neural Network 50. Predictions are aggregated using a bagging technique to maximise performance stability and accuracy. Our results indicate that the Vision Transformers+ Residual Neural Network 50 model achieved the highest accuracy at 98%, significantly outperforming standalone Vision Transformers models with an accuracy of 92.80%. This study demonstrates the potential of integrating Vision Transformers and ResNet models for cardiovascular disease detection, offering a novel and effective solution for improving diagnostic precision in medical imaging.