Enhanced ECG Signals Classification with Image-Driven Ensemble Deep Transfer Learning

Cardiovascular diseases (CVDs) stand as a significant global health concern, characterized by a rising number of cases and fatalities. Within this context, arrhythmias emerge as critical manifestations that require advanced diagnostic tools. In this study, we focus on ECG signals analysis plotted and saved as images employing deep learning methods to address the challenges associated with limited data in medical datasets. Through transfer learning with a relatively small ECG dataset, our research explores the effectiveness of three pre-trained models, inception-V3, DenseNet121, and Xception, in developing robust ECG image classification models. Furthermore, we utilize ensemble learning techniques, specifically averaging, to enhance classification accuracy. Our evaluation results present individual model accuracies and an enhanced weighted average ensemble accuracy of 97.82%, providing a comprehensive and effective approach for arrhythmia detection.