Bindu, G. (57212849033); Swamidason, Iwin Thanakumar Joseph (57742545100); Selvakumar, Mahendran (60206047800); Kiruthika, M. (57224278560); Sarveshwaran, Velliangiri (57195569083); Selvam, R. (60206047900)G. (57212849033); Swamidason, Iwin Thanakumar Joseph (57742545100); Selvakumar, Mahendran (60206047800); Kiruthika, M. (57224278560); Sarveshwaran, Velliangiri (57195569083); Selvam, R. (60206047900)Bindu2026-02-052026-02-0520259798331555030https://dx.doi.org/10.1109/ICESC65114.2025.11212366https://www.scopus.com/pages/publications/105022597516https://gnanaganga.alliance.edu.in/handle/123456789/9031CVDs) are the important reason of international mortality, highlighting the life-threatening need for timely heart disease recognition. Electrocardiogram (ECG) signals have long been utilized for diagnosing cardiac abnormalities, but the application of deep learning models. This study systematically evaluates various deep learning architectures, including VGG-16, MobileNet, Custom CNN, and Vision Transformer (ViT), for heart disease prediction using ECG image data. Models were trained on a labeled dataset containing ECG images. The performance of each method was assessed. Among the models, the Vision Transformer outperformed others, achieving the highest accuracy ($94.6%) and consistently high precision ($94.1%) and recall ($93.8%). The results highlight the efficiency of the ViT in capturing both local and global patterns in ECG images, offering a promising tool for scalable heart disease prediction. Additionally, a cloud-based deployment framework was developed to facilitate real-time diagnostic support in clinical settings. © 2025 IEEE.enCardiovascular Disease; Convolutional Neural Networks; Deep Learning; ECG Image Classification; Heart Disease Prediction; Vision TransformerConference paper