Impact of ECG Dataset Diversity on Generalization of CNN Model for Detecting QRS Complex

Chandan Karmakar,Ahsan Habib,John Yearwood

doi:10.1109/access.2019.2927726

Abstract

Detection of QRS complexes in electrocardiogram (ECG) signal is crucial for automated cardiac diagnosis. Automated QRS detection has been a research topic for over three decades and several of the traditional QRS detection methods show acceptable detection accuracy, however, the applicability of these methods beyond their study-specific databases was not explored. The non-stationary nature of ECG and signal variance of intra and inter-patient recordings impose significant challenges on single QRS detectors to achieve reasonable performance. In real life, a promising QRS detector may be expected to achieve acceptable accuracy over diverse ECG recordings and, thus, investigation of the model's generalization capability is crucial. This paper investigates the generalization capability of convolutional neural network (CNN) based-models from intra (subject wise leave-one-out and five-fold cross validation) and inter-database (training with single and multiple databases) points-of-view over three publicly available ECG databases, namely MIT-BIH Arrhythmia, INCART, and QT. Leave-one-out test accuracy reports 99.22%, 97.13%, and 96.25% for these databases accordingly and inter-database tests report more than 90% accuracy with the single exception of INCART. The performance variation reveals the fact that a CNN model's generalization capability does not increase simply by adding more training samples, rather the inclusion of samples from a diverse range of subjects is necessary for reasonable QRS detection accuracy.

Highlights

Electrocardiogram (ECG) records the bio-electric response of heart’s beating and characterizes a normal heart beat using a P wave, a QRS-complex and a T wave
The main contribution of this study is to investigate the impact of ECG dataset diversity on generalization of a convolutional neural network (CNN) model, both intra and inter-database testing approaches using three publicly available datasets were applied
The main focus of this study is to investigate the generalization capability of a CNN model and not to find the best CNN model for QRS detection from ECG signals, no existing CNN model from literature, for the scope of this study, could be selected and the only option left was to create a new CNN architecture which performs at least, as good as the above CNN models

Summary

Introduction

Electrocardiogram (ECG) records the bio-electric response of heart’s beating and characterizes a normal heart beat using a P wave, a QRS-complex and a T wave. The distinguishing shape of the QRS-complex forms the basis of ECG analysis [1], [2]. Detection of the QRS-complex may trigger the automated analysis of ECG characteristics (i.e., locate neighboring P and T waves, determination of R-R intervals, and heart rate), detection of cardiac anomalies [3], and classification of beats. ECG signal may characterize individual subjects to form unique bio-metric signatures [4]. Over the last three decades, much research has been done on automated QRS detection. The challenges including the non-stationary nature of ECG, presence of

Objectives

Methods

Results

Discussion

Conclusion