Multiclass Signal Quality Assessment of Electrocardiogram using Entropy-based Features and Machine Learning Technique

Electrocardiology (ECG) is useful for deriving a multitude of cardiac functions, and widely used for ambulatory health monitoring. The main hindrance for automatic measurement utilizing ECG is signal quality which gets affected with various artifacts like patient motion, poor skin-electrode contact. While various of ECG signal quality assessment (SQA) techniques are available, most of the works use computationally intensive and memory demanding techniques like wavelet, autoencoders, mode decomposition techniques etc. This paper exhaustively uses various entropy features, which are information rich, computationally lightweight and promising for the application. Secondly, evidence of multiclass ECG SQA is occasional. This paper describes the use of 20 entropy features for 3-class ECG SQA. Total 400 records from Computing in Cardiology (CinC) 2011 challenge, each of 6 leads and of 10 s duration per lead was evaluated, with a blind test accuracy of 75.43% for 3 class and 97% for 2-class SQA. It was found that the entropy features outperform the wavelet and AE features, and random forest provides superior performance over support vector machine and k-NN. The present research provides improved results over few published works on ECG SQA using CinC2011 dataset.