Abstract:Introduction: Deep learning models for detecting episodes of atrial fibrillation (AF) using rhythm information in long-term, ambulatory ECG recordings have shown high performance. However, the rhythm-based approach does not take advantage of the morphological information conveyed by the different ECG waveforms, particularly the f-waves. As a result, the performance of such models may be inherently limited. Methods: To address this limitation, we have developed a deep learning model, named RawECGNet, to detect episodes of AF and atrial flutter (AFl) using the raw, single-lead ECG. We compare the generalization performance of RawECGNet on two external data sets that account for distribution shifts in geography, ethnicity, and lead position. RawECGNet is further benchmarked against a state-of-the-art deep learning model, named ArNet2, which utilizes rhythm information as input. Results: Using RawECGNet, the results for the different leads in the external test sets in terms of the F1 score were 0.91--0.94 in RBDB and 0.93 in SHDB, compared to 0.89--0.91 in RBDB and 0.91 in SHDB for ArNet2. The results highlight RawECGNet as a high-performance, generalizable algorithm for detection of AF and AFl episodes, exploiting information on both rhythm and morphology.
Abstract:Introduction: The presence of fibrillatory waves (f-waves) is important in the diagnosis of atrial fibrillation (AF), which has motivated the development of methods for f-wave extraction. We propose a novel approach to benchmarking methods designed for single-lead ECG analysis, building on the hypothesis that better-performing AF classification using features computed from the extracted f-waves implies better-performing extraction. The approach is well-suited for processing large Holter data sets annotated with respect to the presence of AF. Methods: Three data sets with a total of 300 two- or three-lead Holter recordings, performed in the USA, Israel and Japan, were used as well as a simulated single-lead data set. Four existing extraction methods based on either average beat subtraction or principal component analysis (PCA) were evaluated. A random forest classifier was used for window-based AF classification. Performance was measured by the area under the receiver operating characteristic (AUROC). Results: The best performance was found for PCA-based extraction, resulting in AUROCs in the ranges 0.77--0.83, 0.62--0.78, and 0.87--0.89 for the data sets from USA, Israel, and Japan, respectively, when analyzed across leads; the AUROC of the simulated single-lead, noisy data set was 0.98. Conclusions: This study provides a novel approach to evaluating the performance of f-wave extraction methods, offering the advantage of not using ground truth f-waves for evaluation, thus being able to leverage real data sets for evaluation. The code is open source (following publication).