Machine Learning-Based Automatic Cardiovascular Disease Diagnosis Using Two ECG Leads

Abstract

The state-of-the-art cardiovascular disease diagnosis techniques use machine-learning algorithms based on feature extraction and classification. In this work, in contrast to a conventional single Electrocardiogram (ECG) lead, two leads are used, and autoregressive (AR) coefficients and statistical parameters are extracted to be used as features. Four machine-learning classifiers support-vector-machine (SVM), K-nearest neighbors (KNN), multi-layer perceptron (MLP), and Naive Bayes are applied on these features to test the accuracy of each classifier. For simulation, data is collected from the MIT-BIH and Shaoxing Peoples Hospital China (SPHC) database. To test the generalization ability of our proposed methodology machine-learning model is built on the SPHC database and tested on the MIT-BIH database and self-collected datasets. In the single-database simulation, the MLP performs better than the other three classifiers. While in the cross-database simulation, the SVM-based model trained by the SPHC database shows superiority. For normal and LBBB heartbeats, the predicted recall respectively reaches 100% and 98.4%. Simulation results show that the performance of our proposed methodology is better than the state-of-the-art techniques for the same database. While for cross-database simulation, the results are promising too. Finally, in the demonstration of our realized system, all heartbeats collected from healthy people are classified as normal beats.