Abnormality Detection in ECG Signal applying Poincare and Entropy-based Approaches

Abstract

Detection of abnormality in heart is of major importance for early and appropriate clinical medication. In this work, we have proposed two models for detection of abnormality in ECG signals. The normal ECG signals are closely repetitive in nature to a large extent, whereas ECG signals with abnormalities tend to differ from cycle to cycle. Hence, repetitive plot like the Poincare is efficient to detect such non-repetitiveness of the signal; thereby, indicating abnormalities. Hence, we have used Poincare plot to develop the two proposed models. One of the models uses direct analysis of the binary image of the plot to detect the difference in retracing, between the healthy and unhealthy samples. The other model uses entropy of the Poincare plot to detect the difference in randomness of plots between the two classes. Most importantly, we have used only lead II ECG signal for analysis. This ensures ease of computation as it uses signal of only a single lead instead of the 12 leads of the complete ECG signal. We have validated the proposed models using ECG signals from the ‘ptb database’. We have observed that the entropy analysis of the Poincare plots gives the best results with 90% accuracy of abnormality detection. This high accuracy of classification, combined with less computational burden enables its practical implementation for the development of a real life abnormality detection scheme