Supervised ECG wave segmentation using convolutional LSTM

Abstract

Electrocardiogram (ECG) is the graphical representation of electrical activity of the heart and is used to detect certain structural and functional heart conditions. Segmenting ECG waveforms and annotating constituent components is required for analysis of ECG and to arrive at a diagnosis. This paper proposes a Convolutional Long Short-Term Memory (ConvLSTM) neural network to segment the ECG waves. It consists of a convolutional layer followed by a Bidirectional LSTM architecture. The segmentation is achieved by adding additional features such as derivative of the ECG wave as well as the smoothened ECG wave and the model outperforms traditional Markov models.