ECG beat quality assessment using Self Organizing Map

Abstract

Electrocardiograms (ECG) are cardiac signals formed by the cyclical electrical activity of the heart muscles. The signal is very important for cardiovascular disease assessment. However, in ambulatory monitoring, artifacts are generated quite often which corrupts the ECG beats. This makes the efficient analysis of the biomedical signal difficult. Hence ECG signals needs to be preprocessed efficiently before its medical analysis. This paper describes an approach for preprocessing the noisy ECG signal with the use of Artificial Neural Network (ANN). For this work Self Organizing Map (SOM) of ANN is used. SOM is an unsupervised classifier which automatically distinguishes input data in groups based on the similarities in terms of features. Here various time domain features of ECG were provided as an input to SOM to form a classifier that classifies defective beats from normal beats. For this study, short duration ECG data from the physionet Mitdb database were used. The beats were segmented based on foot detection and the fiducial points were extracted. Some beats of ECG data were mixed with noise collected from the accelerometer. Based on the fiducial points, beat wise features were extracted which were fed to the input of SOM. The SOM will then classify the beats according to their morphology. Similar looking beats will stay in one group. So, by this method it can be easier to classify corrupted beats from the ECG signal. The correct classification of normal and abnormal beats was visually estimated from the weight distance plot of SOM.