A Silent Cardiac Atrial Fibrillation Detection and Classification using Deep Learning Approach

Abstract

The electrocardiogram (ECG is a standard method for diagnosing irregular heart rhythms. Abnormalities, such as silent cardiac atrial fibrillation, which is caused by an irregular cardiac cycle, are detected with the aid of ECG signal data. Faster and more accurate results from automated classification and detection of the ECG arrhythmia signal are considered essential. Improvements in model speed and robustness have been achieved through the application of various pre-processing techniques and deep learning abilities. Many researchers have paid attention to the performance of various deep learning approaches on different datasets of ECG signals. But they have overlooked the significance of data pre-processing before feeding it to deep learning models. This research proposes a Residual Network (ResNet) architecture that increases training stability using a combination of resampling and data augmentation techniques. The results have proven that ResNet produces higher accuracy on the PhysioNet MIT-BIH Arrhythmia dataset for the classification of ECG data.