Cross-domain U-Net framework for unsupervised ECG delineation

The heartbeat in an electrocardiogram (ECG) serves as the primary basis for diagnosing arrhythmias, consisting of P waves, QRS complexes, and T waves. ECG delineation aims to accurately identify and locate these waveforms, providing support for clinical applications such as the diagnosis of heart diseases. However, due to individual differences among patients or variations in data collection equipment, domain shift often occurs in different datasets, resulting in decreased performance of ECG delineation algorithms. To address this issue, this paper proposes an unsupervised domain adaptation delineation model called DA-Unet. This model first augments the task view using a noise-based data augmentation module and then utilizes a domain adaptation module to narrow the gap between the source and target domain feature spaces, thereby enhancing the model's generalization ability in delineating the target domain. Experimental results demonstrate that this approach effectively improves the cross-dataset performance of the delineation model on two public datasets, QTDB and LUDB.