A Deep Learning Framework for the Classification of ECG Signals

The electrical activity of human heart is captured by an electrocardiogram (ECG) signal. The careful analysis of ECG signal can provide valuable insights into the cardiac activity, which can lead to timely therapeutic interventions. The rise of deep learning and the large amount of complex healthcare datasets provides an opportunity to develop intelligent data analysis systems, that can help identify disease patterns and clinical decision support. In this work, we propose a deep neural network model to classify real time ECG signals into normal, atrial fibrillation, and ventricular fibrillation. In the preprocessing stage, the signal is passed through various filters to remove different types of artifacts. The proposed model is trained on a public dataset, containing 30,450 training examples and validated across 6,525 examples and then tested with 6,525 examples. The model achieves testing accuracy of 0.9969%.