Automatic Diagnosis of Atrial Fibrillation Based on Tunable Q-Factor Wavelet Transform and Artificial Neural Networks: Analysis of 12-Lead ECG Signals

Atrial fibrillation (AF) is one of the most common cardiovascular disorders. This condition, which can occur due to various reasons, seriously affects the quality of life of people when it is not diagnosed early. One of the most important techniques used in the diagnosis of AF is the electrocardiogram (ECG). ECG signals consist of different waveforms, which are analyzed by cardiologists to diagnose the disease. However, in some cases, these waveforms can be difficult to identify and can make disease diagnosis difficult. At this point, artificial intelligence-based computer-aided diagnosis (CAD) systems can help physicians. With such automatic diagnostic systems, AF attacks can be prevented quickly and early and the workload of physicians can be reduced. In order to develop such systems, this research deals with the analysis and classification of 12-lead ECG signals of AF patients and healthy control (HC) groups using machine learning methods. Tunable Q-factor wavelet transform (TQWT) is applied to the ECG signals and the features that will significantly affect the classification performance are determined by the Relief method and these features are analyzed with three different machine learning classification models. The experimental results show that high accuracy, specificity, sensitivity, precision, accuracy, AUC, and F1-Score (96% and above) are obtained in performance metrics, especially when Artificial Neural Networks (ANN) algorithm is used. This research reveals that TQWT-ANN-based systems can help in the automatic diagnosis of AF.