Abstract
We use the Filter Diagonalization Method (FDM), a harmonic inversion technique, to extract f-wave features in electrocardiographic (ECG) traces for atrial fibrillation (AF) stratification. The FDM detects f-wave frequencies and amplitudes at frame sizes of 0.15 seconds. We demonstrate our method on a dataset comprising of ECG recordings from 23 patients (61.65 ± 11.63 years, 78.26% male) before cryoablation; 2 paroxysmal AF, 16 early persistent AF (<; 12 months duration), and 4 longstanding persistent AF (>12 months duration). Moreover, some of these patients received adenosine to enhance their RR intervals before ablation. Our method extracts features from FDM outputs to train statistical machine learning classifiers. Tenfold cross-validation demonstrates that the Random Forest and Decision Tree models performed best for the pre-ablation without and with adenosine datasets, with accuracy 60.89 ± 0.31% and 59.58% ± 0.04%, respectively. While the results are modest, they demonstrate that f-wave features can be used for AF stratification. The accuracies are similar for the two tests, slightly better for the case without adenosine, showing that the FDM can successfully model short f-waves without the need to concatenate f-wave sequences or adenosine to elongate RR intervals.
Original language | English |
---|---|
Title of host publication | 2019 Computing in Cardiology (CinC) |
Place of Publication | Singapore |
Publisher | IEEE Xplore |
ISBN (Electronic) | 978-1-7281-6936-1 |
ISBN (Print) | 978-1-7281-5942-3 |
Publication status | Published - 24 Feb 2020 |
Keywords
- feature extraction
- electrocardiography
- decision trees
- atrial fibrillation
- arrhythmia
- disease stratification
- filter diagonalisation method
- fibrillatory waves