Transfer Learning Based Paroxysmal Atrial Fibrillation Classification Using Continuous Wavelet Transform Scalogram

Atrial fibrillation (AF), the most common cardiac arrhythmia, is linked to comorbidity, heart failure, and aging. Its erratic and intermittent behavior makes reliable detection difficult. Recent research has demonstrated the usefulness of nonlinear time and frequency domain characteristics in the detection of AF through the application of deep learning (DL) and other machine learning (ML) techniques. In this study, the CPSC 2021 dataset was used to create continuous wavelet transform (CWT)-based scalogram images from cardiac rhythms comprising normal, persistent atrial fibrillation (PAF), and paroxysmal atrial fibrillation (PoxyAF). The challenge involved classifying normal and PAF scalogram images, as well as three classes of classification: normal, PAF, and PoxyAF, for which transfer learning (TL) models were utilized. A number of TL models were trained; among them, EfficientNetB7 obtained a testing accuracy of 99.27% for two-class classification, whereas MobileNet achieved 93.23% for three-class classification.