Do Single Beat ECGs have Same Credibility as Continuous ECG Signal in Multi-Class Classification using Machine Learning Models

Artificial intelligence has proven to be a boon for humanity in various field of science, however meaningful results using Machine Learning (ML) and Deep Learning (DL) models depends on various factors. One of the factors is to feed the model with large size datasets. For this purpose, segmentation of ECG into Single Beat ECGs (SB-ECG) has become prominent. This study attempts to investigate whether there is an effect of using SB-ECG compared to the continuous recorded ECGs on the model output. The aim of this work is to examine and compare the effect of single beat and continuous ECGs on the ML model multi-class classification performance. 120 ECGs of Sinus Rhythm (SR), Sinus Tachycardia (ST) and Atrial Tachycardia (AT) recorded for 10 seconds using standard 12 lead were retrieved from online database. Two datasets were prepared of 120 and 1687 samples of continuous ECG and SBECG, respectively. The datasets were fed to the ML model and their performance were evaluated. Extra trees model gave the best performance for both datasets. Continuous ECG recording gave the 10 -fold cross validated data accuracy of 77.36% and SB-ECGs gave 92.97% with Area Under Curve (AUC) score of 88.07 and 98.92 respectively. SB-ECG gave better classification performance than the continuous ECG recording. SB-ECG can be used for enhanced clinical diagnosis with better accuracy and reduced false outcomes. Also, the study on SB-ECG may prove to be highly efficient in paroxysmal atrial fibrillation detection where episodes of arrhythmia are sudden.