Physiology-Informed ECG Delineation Based on Peak Prominence

Reliable wave morphology extraction from the electrocardiogram (ECG) is crucial for enabling effective and trustworthy diagnosis and monitoring of heart conditions. Therefore, we propose a novel physiology-informed ECG delineation algorithm based on prominence information as an explainable metric for morphology point selection. Key advantages are linear runtime complexity, physiology-based parameter choices, and the possibility of enabling real-time waveform annotation within heartbeats in streaming ECG data. Additionally, the proposed method benefits from a multi-lead correction step, enhancing delineation performance when data from multiple leads is available. Evaluations on two ECG databases showcase the superior performance of the proposed prominence delineator compared to established methods. In particular, F1-scores above 99% are achieved for nearly all morphology waves on the Lobachevsky University Database (LUDB) and QT Database (QTDB), with the only exception of the T-wave on the QTDB. Moreover, standard deviations of detection errors lie significantly below the tolerances set by the Committee of General Standards for Electrocardiography. Thus, our physiology-informed unsupervised signal processing approach concurrently increases performance and explainability compared to the state of the art. An open source implementation is provided to facilitate practitioners to conduct explainable and reliable ECG-based biomarker identification for detecting heart conditions.