PaperlayerView exact revision on arXiv

Source-linked AI summary

Trajectory Analysis of ECG Motif Dynamics in the Run-up to Sudden Cardiac Arrest

Nivedita Bijlani, Mauricio Villarroel

arXiv:2609.01543v1eess.SP

TL;DR

Early warning signatures of SCA are poorly characterised in long-duration ECG, where existing analysis remains largely event-centric. The paper uses annotation-free motif trajectories to quantify patient-specific morphological change in 23 Holter recordings, finding sustained abnormalities hours before VF, with dispersion strongest for long-horizon detection and consistency strongest near onset.

  • Problem

    Early warning signatures of SCA are poorly characterised, and event-centric ECG analysis provides limited tracking of evolving morphology.

    Source: §1 · p. 1

  • Method

    The study extracts representative motifs from 10 s ECG windows, builds trajectories across five morphology metrics, and normalises them against an early personalised baseline.

    Source: §1 · p. 1, §2.2 · p. 2, §2.3 · p. 2

  • Results

    5.8–8.4 h before VF, median sustained abnormal-burden onset occurred; motif dispersion detected burden at least 1 h before VF in 100% and at least 2 h in 89%.

    Source: §3 · p. 3, §4 · p. 4

  • Takeaways & Limitations

    The framework provides an interpretable, label-free representation of evolving cardiac morphology for continuous long-duration ECG monitoring.

    Source: §4 · p. 4, §5 · p. 4

    This retrospective study provides exploratory evidence, and sensitivity to alternative analysis settings remains unevaluated.

    Source: §2.5 · p. 3, §4 · p. 4

Abstract

from arXiv · show

Early warning signatures of sudden cardiac arrest (SCA) remain poorly characterised in long-duration ECG. We quantified pre-event changes in ECG morphology using a motif-based trajectory framework. Holter ECGs from 23 patients with annotated SCA were analysed over non-overlapping 10 s windows. Window-level motifs were extracted to quantify trajectories of instability, consistency, dispersion, heterogeneity, and personalised-baseline distance. Each trajectory was normalised to an early baseline using z-scores and aligned to ventricular fibrillation (VF) onset. Abnormal burden was defined as the proportion of windows with z>=3 within a rolling 10-minute window. Median sustained abnormal burden onset occurred 5.8-8.4 h before VF across morphological metrics. Motif dispersion showed the most consistent long-horizon detection, with sustained abnormal burden >=1 h before VF in 100% of patients and >=2 h in 89%. Motif consistency showed the strongest late-stage change, with 70% median abnormal burden in the final 10 min. Peak morphological deviation occurred ~2 h prior to VF. Our label-free framework transforms longitudinal ECG analysis from an event detection approach towards a continuous characterisation of evolving cardiac change, enabling a personalised early warning of SCA, well suited to long-duration wearable ECG monitoring.

1. Introduction

Sudden cardiac arrest risk is difficult to identify early, despite disease-related electrophysiological and morphological changes that may evolve before the event. The paper addresses the gap with continuous, interpretable, annotation-free tracking of patient-specific ECG morphology.

  • SCA remains difficult to predict before onset, although disease substrates manifest as changes in cardiac electrophysiology and ECG morphology.Source: §1 · p. 1
  • Long-duration ECG systems are largely event-centric and provide limited information about how morphology evolves before adverse events.Source: §1 · p. 1
  • Representative ECG motifs construct continuous signatures of patient-specific morphological change, stability, and variability without predefined event labels.Source: §1 · p. 1

2. Methods

The study builds sequential motif trajectories from 10-second ECG windows, quantifies complementary morphology metrics, and normalises them against personalised early baselines. Abnormal burden is then used to identify sustained deviations across multiple warning horizons.

2.1. Dataset

  • The framework summarises longitudinal ECG morphology through representative motifs extracted from consecutive windows.Source: §2 · p. 1, §2.3 · p. 2
  • 23 two-channel Holter recordings from the Sudden Cardiac Death Holter Database were analysed, with pre-VF ECG used for personalised baselines where available.Source: §2.1 · pp. 1–2

2.2. ECG preprocessing

  • ECGs were divided into consecutive 10 s windows, and R-peak-centred segments were temporally normalised to 650 samples before motif extraction.Source: §2.2 · p. 2

2.3. Motif extraction

  • The representative motif was selected as the segment with the lowest median Dynamic Time Warping distance to all other segments in its window.Source: §2.3 · p. 2

2.4. Motif-based morphological metrics

  • Five metrics captured baseline distance, consistency, dispersion, heterogeneity, and instability within or across windows.Source: §2.4 · p. 2

2.5. Trajectory and early-warning analysis

  • Trajectories were aligned to VF onset, normalised to the first 15 min baseline, and thresholded at z ≥3 to compute rolling 10 min abnormal burden.Source: §2.5 · pp. 2–3
  • Burden onset required at least 40% abnormal windows sustained for 2 min, with early warning assessed at 30, 60, and 120 min before VF.Source: §2.5 · p. 3
  • Thresholds were selected a priori to capture marked, sustained deviations while limiting sensitivity to transient fluctuations.Source: §2.5 · p. 3

3. Results

Motif trajectories showed sustained morphological changes hours before VF, with dispersion providing the most consistent long-horizon detection and consistency showing the strongest final-stage change.

  • 5.8–8.4 h before VF, median abnormal-burden onset occurred across the five motif metrics.Source: §3 · p. 3
  • 93–100% of recordings showed sustained abnormal burden at least 30 min before VF, versus 88–100% at 1 h and 67–89% at 2 h.Source: §3 · p. 3
  • Maximum slope occurred 1.7–3.4 h before VF and peak deviation 1.9–2.2 h before VF.Source: §3 · p. 3

4. Discussion

Motif-based ECG trajectories captured progressive pre-VF morphological change across complementary temporal patterns, with sustained abnormalities emerging hours before VF. The findings support continuous, patient-specific monitoring while remaining exploratory pending prospective validation.

  • 8.4 and 7.7 h before VF marked the median onset of sustained abnormal burden for within-window consistency and dispersion, respectively.Source: §4 · p. 4
  • 100% of patients showed sustained abnormal dispersion burden at ≥1 h before VF, declining to 89% at ≥2 h.Source: §4 · p. 4
  • 70% median abnormal burden for consistency in the final 10 min represented the strongest late-stage change.Source: §4 · p. 4
  • Approximately 2 h before VF, instability and personalised-baseline distance reached peak deviations after sustained changes several hours earlier.Source: §4 · p. 4
  • The framework quantifies patient-specific morphology without predefined rhythm labels or event-labelled training data, using compact motifs linked directly to the ECG.Source: §4 · p. 4
  • Prospective evaluation in larger cohorts is still needed, and sensitivity to alternative analysis settings remains unevaluated.Source: §4 · p. 4

5. Conclusion

The study transforms representative ECG motifs into continuous, interpretable signatures of patient-specific cardiac morphology before VF. This lightweight, annotation-free approach could support earlier assessment when integrated into long-duration wearable ECG monitoring.

  • Representative ECG motifs were transformed into continuous longitudinal signatures that quantify patient-specific morphological change before VF.Source: §5 · p. 4
  • By modelling deviation, stability, and variability over time rather than discrete events, the approach provides an interpretable, annotation-free representation of evolving cardiac state.Source: §5 · p. 4
  • Integrated into wearable ECG, lightweight analysis could create an actionable window for earlier clinical assessment and intervention.Source: §5 · p. 4
Loading 2609.01543v1…