[論文レビュー] A novel method for analysis of transient morphological changes in quasiperiodic physiological signals and their neurogenic correlates

Tomasz Gradowski, Damian Waląg+2 more|arXiv (Cornell University)|Feb 22, 2026

ECG Monitoring and Analysis被引用数 0

ひとこと要約

Introduces carpet plots to visualize quasiperiodic signals by aligning cycles at rhythmic triggers, enabling simultaneous analysis of rhythm and morphology in ECG and other signals, with potential for AI-assisted analysis.

ABSTRACT

Frequently, transient changes in physiological signals, such as ECG morphology, precede or follow a rate change. Current methods for visualizing morphology allow only the tracking of preselected changes, severely limiting analytical capabilities. We introduce a novel method for visualizing quasiperiodic signals, enabling the transformation of time series containing repetitive patterns into intuitive visual representations. By using segmentation algorithms and color encoding, we generate two-dimensional "carpet plots" that facilitate simultaneous assessment of heart rhythm and signal features, including the morphology of QRS complexes and T waves, as well as transient changes in intervals and amplitudes. Additionally, the method supports the assessment of concomitant changes in morphology and rate. Typically, existing visualization methods, such as the standard 12-lead ECG projection, focus either on rhythm variability or on morphological analysis of a few consecutive beats. In contrast, our method integrates both aspects into a single, coherent graphical representation, greatly enhancing the detection of subtle disturbances and a fascinating dynamic interplay between the rhythm and the morphology of the signal. We illustrate the effectiveness of this approach using Holter recordings from healthy individuals and patients with arrhythmias, as well as stress test sessions. The results highlight the potential of our visualization technique to support diagnosis and long-term ECG signal analysis. The method may be applied to a broad class of repeatable quasiperiodic patterns - we demonstrate a few examples.

研究の動機と目的

• Motivate joint analysis of rhythmic and morphological features in quasiperiodic physiological signals.
• Develop a visualization method that captures intra- and inter-cycle variability in morphology across cycles.
• Enable assessment of coupling between rhythm and morphology and support long-term ECG analysis.
• Demonstrate applicability to ECG and extendable applicability to multimodal signals.

提案手法

• Identify a characteristic feature of quasiperiodic patterns and segment the signal around central pattern occurrences.
• Align segments by a central pattern (e.g., R peaks) to create uniform-length windows.
• Map segment amplitudes to color values using a transfer function for visualization (amplitude-to-color mapping).
• Stack segments chronologically to form a two-dimensional carpet plot with cycles on the vertical axis and relative time within a cycle on the horizontal axis.
• Use fixed segment duration and resample signals to a common rate to ensure uniform carpet plot dimensions for ML compatibility.
• Discuss preprocessing choices like baseline handling, amplitude clipping, and color scale adjustments to improve readability.

実験結果

リサーチクエスチョン

• RQ1 How can joint rhythm and morphology be visualized to reveal their coupling over time?
• RQ2 Can carpet plots detect subtle morphological changes and phase transitions in quasiperiodic signals?
• RQ3 How can this visualization be leveraged to support automated analysis and ML-based pattern detection in ECG and multimodal signals?

主な発見

• Carpet plots enable simultaneous visualization of heart-rate dynamics and ECG morphology across long recordings.
• The method reveals phase transitions and transient episodes such as atrial fibrillation, AV-block patterns, and ST-segment changes.
• Color mapping and axis alignment facilitate detection of subtle changes in QRS duration, QT intervals, and T-wave morphology.
• ResNet18 feature maps applied to carpet plots demonstrate extractable, interpretable patterns suitable for CNN-based analysis.
• The approach is adaptable to multimodal signals (ECG, ABP, ICP) by decoupling the segmentation feature from the target signal.