It is our pleasure to annonce the publication of our study entitled “Machine learning for ranking f-wave extraction methods in single-lead ECGs” in Biomedical Signal Processing and Control. The study was led by Noam Ben-Moshe and in collaboration with Shany Biton Brimer, Kenta Tsutsui, Mahmoud Suleiman, Leif Sörnmo and Joachim A. Behar
Illustration of f-wave extraction. (a) A single-lead ECG with AF and related f-waves extracted using (b) ABS, (c) ABSsc1, (d) ABSsc2, and (e) TSPCA. In this example, TSPCA is associated with the smallest QRS-related residuals.
The presence of fibrillatory waves (f-waves) is important in the diagnosis of atrial fibrillation (AF), having motivated the development of several methods for f-wave extraction. This study presents a novel method for assessing the performance of single lead f-wave extraction algorithms, offering the distinct advantage of evaluating those algorithms on real data sets without relying on ground truth f-waves for evaluation. The new methods rely on the hypothesis that better-performing AF classification, using features computed from the extracted f-waves, implies better-performing extraction.
We demonstrate the validity of the new method on 300 manually annotated Holter recordings, totaling >7000 hours of continuous ECG data from three geographical locations as well as simulated ECG data. Four standard extraction methods based on either average beat subtraction or principal component analysis (PCA) were benchmarked. The PCA based method consistently performed best across real data sets and lead position. On the simulated data set, it also performed best as well as demonstrated the lowest RMS error residual inside and outside the QRS interval.
Overall, this research provides a new methodology, and associated open-source code, for assessing the performance of single lead f-wave extraction algorithms. The robust extraction of the f-wave can enable more advanced research in its characteristics and association with clinical outcomes.
Open source code: https://github.com/noambenmoshe/fwave
Link to the manuscript: https://www.sciencedirect.com/science/article/abs/pii/S1746809424008759
Link to one of the dataset which we opened: https://physionet.org/content/shdb-af/1.0.0/
AIMLab. https://www.aimlab-technion.com/
コメント