Quality assessment of ECG signals has crucial importance for the automatic diagnosis of heart diseases, especially for signals which are heavily contaminated with several artifacts. Therefore, several SQA (signal quality assessment) techniques were presented based on ECG signal features and the machine learning classifiers or heuristic decision rules. This study presents an algorithm for accurate and offline detection of motion artifacts and noise in ECG signals. Our grading algorithm involves two stages. The first stage involves extracting several non-fiducial features from ECG. The second stage of our approach uses Support Vector Machines and Binary Decision Trees to grade signals. The proposed method was tested using the PhysioNet/Computing in Cardiology Challenge 2011 Database by comparing two different classifiers. The SVM provided the best classification accuracies of nearly 94% on the labeled data. (Labels which indicate the ECG signal is acceptable or not for clinical interpretation).
Paper
Full text
Quality Assessment of ECG Signals Based on Support Vector Machines and Binary Decision Trees
Semantic Scholar · Medicine · 2020
Abstract
Quality assessment of ECG signals has crucial importance for the automatic diagnosis of heart diseases, especially for signals which are heavily contaminated with several artifacts. Therefore, several SQA (signal quality assessment) techniques were presented based on ECG signal features and the machine learning classifiers or heuristic decision rules. This study presents an algorithm for accurate and offline detection of motion artifacts and noise in ECG signals. Our grading algorithm involves two stages. The first stage involves extracting several non-fiducial features from ECG. The second stage of our approach uses Support Vector Machines and Binary Decision Trees to grade signals. The proposed method was tested using the PhysioNet/Computing in Cardiology Challenge 2011 Database by comparing two different classifiers. The SVM provided the best classification accuracies of nearly 94% on the labeled data. (Labels which indicate the ECG signal is acceptable or not for clinical interpretation).