• Version
• Download 12
• File Size 505.64 KB
• File Count 1
• Create Date 20 March 2026
• Last Updated 20 March 2026

Lightweight CNN-Based approach for Multi-Class ECG Image Classification

1 B. Rishika, 2 E. Shivani, 3 Mohammed Nawaz Uddin, 4 K. Aakash, 5 Dr. V. Neelima

Department of CSE (Artificial Intelligence & Machine Learning)

Jyothishmathi Institute of Technology and Science Karimnagar, Telangana, India

rishika.billa04@gmail.com, www.shivanienugu@gmail.com, nawaznayeem2562@gmail.com, aakashking147@gmail.com

Abstract— Cardiovascular diseases (CVDs) continue to be the highest cause of mortality worldwide.[1] This has necessitated the development of efficient, automated, and accurate diag-nostic techniques. Although traditional Convolutional Neural Networks (CNNs) perform well in image recognition, there is limited efficiency in dealing with specific multi-scale depen-dencies of Electrocardiogram (ECG) signals. The purpose of this research is to propose a Parallel Branch Deep Fusion CNN model for diagnosing four different states of cardiac conditions, including Normal, Abnormal Heartbeat, Myocardial Infarction (MI), and History of MI. This model is based on a two-stream architecture, including a Spatial Branch, which employs a pre-trained ResNet-18 model, and a Global Context Branch, designed to obtain long-range dependencies of signals. The model is then optimized using a phased learning mechanism. The proposed model was found to attain a maximum validation accuracy of 89.06%, indicating better stability compared to traditional CNN architectures.