Classification of Cardiac Abnormalities using Deep Learning and Machine Learning Methods

Abstract

Cardiovascular disease detection through ECG image analysis presents a critical challenge in modern healthcare diagnostics. This research introduces an innovative approach combining SqueezeNet architecture with advanced data augmentation techniques for automated cardiac abnormality classification. Our study analyzes a comprehensive dataset of ECG images representing four distinct cardiac conditions: normal rhythm, abnormal heartbeat, myocardial infarction, and historical myocardial infarction cases. We hyperparameter tuning; second, comparing its performance with AlexNet and a lightweight CNN; and third, utilizing these architectures charcteristic extractors for traditional machine learning classifiers. The optimized SqueezeNet model achieves remarkable results with 99.85% accuracy, 99.7% precision, 99.8% recall, and 99.8% F1-score, significantly outperforming existing approaches. Notable improvements include enhanced feature extraction capabilities, demonstrated by achieving 99.85% accuracy when combined with the SVM classifier. The model's efficiency is particularly evident in its reduced computational requirements, made it suitable for deployment on resource restraints devices.