ECG Heartbeat Segmentation with Bidirectional LSTM Deep Learning Neural Network

Abstract

The electrocardiogram (ECG) is the most representative signal to be used and illustrated for exactly understanding electrical activity of the heartbeat. Most of the heart diseases can be detected by means of ECG signal which consists of multiple heartbeat malfunctional types. Each heart disease may consist of different types of heartbeat samples. In general, these heartbeat samples could be segmented by physician annotations. However, in addition to need more time to judge ECG signal, physicians even with more than 10 years of clinical experience may be difficult to give an accurate ECG judgment. Therefore, an automated ECG heartbeat segmentation is required for diagnosis and analysis of heart disease. In our research, we employed python programming language with its diverse workable modules to build Bidirectional Long Short-Term Memory (BiLSTM) neural networks model to distinguish seven databases from the QT database. The QT databases could be obtained from open-source PhysioNet. These seven disease databases of the QT database are available from the MIT-BIH Arrhythmia Database, the European Society of Cardiology ST-T Database, MIT-BIH ST Change Database, MIT-BIH Supraventricular Arrhythmia Database, MIT-BIH Normal Sinus Rhythm Database, MIT-BIH Long-Term ECG Database, and sudden death database. Finally, 89% heartbeat segmentation accuracy for our research could be obtained to help physicians for ECG heartbeat segmentation.