ECG: Deep Learning for Arrhythmia Classification
Electrocardiography signals are manually interpreted for the diagnosis of cardiac arrhythmias. For efficient screening of arrhythmia from long term ECG data, an automated ECG interpretation is required. However, the existing automated ECG interpretation devices require extensive pre-processing and knowledge to determine relevant features. Hence, there is a need for a comprehensive feature extractor and classifier to analyse ECG signals.
An automated ECG interpretation has traditionally been done with a combination of QRS detection, feature extraction and machine learning classifiers. Access to powerful computing hardware and advancements in deep learning network architectures have resulted in an influx of deep learning networks to diagnose cardiac arrhythmias ECG.
HTIC has proposed three robust deep neural network (DNN) architectures to perform feature extraction and classification of a given two second ECG signal. The first network is a Convolutional Neural Network (CNN) with multiple kernel sizes, the second network is a Long Short Term memory (LSTM) network and the third network is a combination of CNN and LSTM based feature extractor, CLSTM network. The proposed networks are end to end networks which can be directly trained without any pre-processing. The network would also be easily adaptable to multiple datasets requiring minimal training only on the final three layers through use of transfer learning.
The model performance was validated with multiple datasets such as the MITDB, LTDB and LTAFDB arrhythmia databases. The networks were trained and tested with the MITDB ECG dataset on three classes Normal (N), Premature Ventricular Contraction (PVC) and Premature Arterial Contraction (PAC)
The best model CLSTM gave an accuracy of 97.6%. The results showcase the potential of the network as feature extractor for ECG datasets. Our results outperform the state-of-the art works on ECG classification on several metrics.