Publication Type : Conference Paper
Publisher : Procedia Computer Science
Source : Procedia Computer Science (2020)
Url : https://www.sciencedirect.com/science/article/pii/S1877050920310322
Keywords : Bidirectional Long Short-Term Memory Networks, Convolutional neural network, Deep learning, electrocardiogram, Electrophysiology
Campus : Coimbatore
School : School of Engineering
Center : Computational Engineering and Networking
Department : Electronics and Communication
Year : 2020
Abstract : Electrocardiogram (ECG) is a vital tool to determine the structure and function of the heart. Automated classification of the ECG data helps in earlier diagnosis before the doctor could personally investigate. Most of the cardiac diseases are life threatening universally. In an effort towards accurate detection of cardiac disease by automated identification, several dataset has been released [Physionet]. Deep learning requires massive amount of dataset to be able to provide best results, hence we chose to use the dataset from The China Physiological Signal Challenge 2018, that contains 12 lead ECG recordings from 6877 subjects with eight major cardiac anomalies. To improve the detection / prediction of the Electrophysiology anomalies, a data driven approach was employed. Computational Experiment was performed with novel architecture and approach using Deep Neural Network (DNN) that has resulted in relatively very high accuracy of 99.01%. A minimal set of ECG leads to get maximum accuracy was identified.
Cite this Research Publication : S. VG and Dr. Soman K. P., “Towards identifying most important leads for ECG classification. A Data driven approach employing Deep Learning”, in Procedia Computer Science, 2020.