DEVELOPMENT OF AN AI-POWERED ELECTROCARDIOGRAM IMAGE ANALYSIS SYSTEM

ABSTRACT Electrocardiogram (ECG) interpretation is vital for diagnosing cardiovascular diseases (CVDs) and detecting various cardiac abnormalities. However, in many resource-limited healthcare settings, reliance on trained specialists often causes delays, inconsistent interpretations, and potential errors. This study addresses these challenges through the design, development, and evaluation of a web-based, AI-powered diagnostic system that automatically classifies cardiac conditions from ECG images. A hybrid deep learning architecture combining Convolutional Neural Networks (CNNs) for local feature extraction with Vision Transformers (ViTs) for modeling long-range dependencies and capturing the global structure of ECG waveforms was implemented. The model was trained and validated on a diverse set of publicly available ECG images covering three diagnostic categories: Normal, Arrhythmia, and Myocardial Infarction (MI). Data preprocessing and augmentation were applied to improve generalization and minimize overfitting. The system was deployed via a user-friendly web interface that enables healthcare providers to upload ECG images and receive real-time diagnostic predictions. Evaluation results showed that the hybrid model achieved an overall accuracy of 92%, a weighted F1-score of 87%, and an AUC of 0.92, outperforming CNN-only and ViT-only models. By providing rapid, accurate, and accessible diagnostic support, this system can serve as an effective clinical decision support tool. Its deployment could improve early detection of cardiovascular conditions, ease diagnostic workloads for clinicians, and enhance patient outcomes particularly in underserved regions with limited cardiology expertise.