Title: Interpretable ECG Screening and Monitoring via Motif-Based Morphological Signatures

Abstract: Despite the pivotal role of electrocardiography (ECG) in cardiovascular assessment, its interpretation continues to depend heavily on manual, episodic analysis. Standard clinical workflows utilize brief resting ECGs and, when necessary, extended ambulatory recordings; however, the resulting data demand intensive review efforts. This resource-heavy process often allows subtle morphological shifts or progressive drifts—precursors to clinically significant abnormalities—to escape detection. To address this, we present a motif-based framework that establishes beat-aligned ECG motifs as interpretable cardiac signatures, allowing for the quantification of morphological drift and deviation across both short- and long-term monitoring periods. These motifs serve as representative cardiac cycles that capture dominant morphological features. We developed three distinct metrics for interpretability: deviation from normal sinus rhythm (NSR), deviation from a personalized baseline, and a motif instability index. Motif extraction is performed by identifying beats that minimize Dynamic Time Warping (DTW) distance within fixed time windows. Our evaluation utilized the PTB-XL dataset for short-term analysis and the MIT-BIH Arrhythmia dataset for long-duration monitoring. Interpretability is facilitated through fiducial-based visualizations and representative motif overlays, which allow for the direct inspection of morphological variations. Results from the MIT-BIH dataset indicate that our proposed metrics effectively differentiate predominantly normal subjects from those with arrhythmias (p<0.01). Furthermore, in the PTB-XL dataset, NSR deviation successfully distinguished normal ECGs from abnormal ones across major diagnostic subtypes (p<1e-4, with Cliff’s delta reaching up to 0.93). By offering an interpretable representation of cardiac morphology, ECG motifs support scalable longitudinal monitoring and the early identification of morphology-driven changes.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC