📄 cardiovascular medicine

FoxTail: An R-Peak-Anchored Event Domain for Visualizing and Quantifying Changes in ECG Dynamics

The paper introduces FOXTAIL, a novel R-peak-anchored event domain that visualizes and quantifies beat-to-beat changes in ECG signal direction to reveal dynamic cardiac organization and state variations that conventional methods often miss, without replacing standard diagnostic ECGs.

Original authors: Garcia, N. M.

Published 2026-08-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Garcia, N. M.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

For over a century, the electrocardiogram has been the standard tool for listening to the heart's electrical rhythm. By tracing voltage changes over time on a piece of paper or a screen, doctors can identify irregular beats, measure the timing of electrical signals, and diagnose many heart conditions. This traditional view is like watching a movie frame by frame; it is excellent for seeing the shape of a single heartbeat or the pattern of a few seconds of rhythm. However, when a doctor needs to understand how the heart changes over hours or days, the standard display has a blind spot. It is difficult to see how the internal shape of a heartbeat evolves from one second to the next when hundreds or thousands of beats are stacked together. The subtle shifts in the electrical path that happen between beats often get lost in the noise of a long recording, hidden by the sheer volume of data.

A new approach called FOXTAIL aims to fill this gap by changing the way we look at the heart's electrical activity. Instead of plotting voltage against time, this method organizes the data based on the direction the signal is moving. Imagine a single heartbeat as a journey that goes up and down. The traditional method records every tiny step of that journey. FOXTAIL, however, ignores the steps that continue in the same direction and only marks the moments where the path turns. It captures the peaks and valleys where the signal reverses course. By lining up hundreds of these "turning points" from consecutive heartbeats, the researchers created a new visual map. This map does not show time passing; it shows the sequence of turns. When many heartbeats are overlaid on this map, the result looks like a fan or a tail, with a dense cluster of turns near the start of the heartbeat and a more scattered, variable pattern as the heartbeat progresses. This structure allows scientists to see how the organization of the heart's electrical path changes from beat to beat, revealing patterns that are invisible in the standard view.

The researchers tested this idea using recordings from several different groups of people, including those with normal heart rhythms, those with severe heart failure, and those experiencing episodes of atrial fibrillation, a condition where the heart beats irregularly. They also tested the method on recordings with added noise to see if the new view could distinguish between a sick heart and a dirty signal. The study found that the FOXTAIL view could indeed reveal changes within a single person's heart that a standard average beat would miss. Specifically, in patients who were about to enter an episode of atrial fibrillation, the pattern of turning points changed in a measurable way. The heartbeats leading up to the event showed a slightly different density of turns and a different level of stability compared to the heartbeats recorded when the patient was stable. This suggests that the heart's electrical organization shifts subtly before a major rhythm disturbance occurs, and the FOXTAIL method can detect that shift.

However, the study also made it clear that a dense or complex-looking pattern does not automatically mean the heart is sick or that a disease is coming. When the researchers added artificial noise to the recordings, the pattern of turns changed, but not in the same way it changed for the patients with heart failure. This proved that a messy-looking graph is not always a sign of a complex biological state; sometimes it is just a sign of a poor signal. The method successfully separated the signal of a changing heart from the signal of a noisy measurement. Furthermore, the researchers looked for a specific warning sign before a dangerous heart rhythm called ventricular fibrillation, but they did not find a single, universal pattern that appeared in every patient right before the event. This negative result is just as important as the positive ones, as it shows that the heart does not always follow a predictable script before a crisis, and that no single visual trick can replace careful clinical judgment.

The core achievement of this work is not a new way to diagnose a disease, but a new way to observe the heart. It offers a complementary view that sits alongside the traditional electrocardiogram, asking a different question: how is the electrical path of the heart changing from one beat to the next, and which of those changes last? By focusing on the sequence of directional turns rather than the raw voltage over time, FOXTAIL makes the evolution of the heart's electrical organization visible. It allows researchers to measure the stability of these changes and see how they persist across different levels of detail. While it is not yet a tool for immediate clinical diagnosis, it provides a powerful new lens for understanding the dynamic nature of the heart, helping scientists distinguish between the subtle shifts of a changing physiological state and the random chaos of measurement noise.

Technical Summary: FoxTail: An R-Peak-Anchored Event Domain for Visualizing and Quantifying Changes in ECG Dynamics

Problem Statement
Conventional electrocardiography (ECG) excels at waveform morphology and rhythm diagnosis but struggles to visualize how the internal shape of hundreds or thousands of consecutive heartbeats evolves over time. Standard displays (scrolling monitors, long strips, or ensemble averages) either limit morphological resolution, fail to show continuous internal trajectory changes, or treat beat-to-beat variation as mere dispersion around a representative beat. While heart-rate variability and nonlinear analyses quantify dynamics, they typically operate on R–R intervals or summary features rather than the ordered, within-cycle changes of the ECG waveform itself. There is a representational gap between inspecting individual morphologies and reducing long recordings to aggregate descriptors.

Methodology
The paper introduces FOXTAIL, an R-peak-anchored event domain that transforms the ECG from a uniformly sampled time series into an ordered sequence of directional events.

Signal Transformation: The method divides the signal into natural, variable-duration R-to-R cycles. Within each cycle, samples that continue a monotonic rise or fall are suppressed. Only samples that reverse the direction of the trajectory (turning vertices) and the cycle endpoints are retained. This creates a sparse sequence of local extrema.
Visual Representations: Two complementary views are generated:
1. Amplitude View: Overlays the retained vertices ( $v_i[k]$ ) without normalizing vertical scale, preserving the relationship to the conventional ECG and R-peak amplitude.
2. Differential View: Overlays the magnitudes of signed differentials between consecutive vertices ( $|\delta_i[k]|$ ), removing baseline offsets to emphasize excursion size.
Descriptors: The method calculates several quantitative descriptors:
- Retention Ratio ( $\rho$ ): The fraction of samples retained as vertices.
- Event Density ( $E$ ): The number of turning events per cycle.
- Dispersion: The standard deviation of retention ratios ( $SD(\rho)$ ) and changes in retention ( $SD(\Delta\rho)$ ) across beats.
- Multiscale Persistence: By applying Gaussian smoothing at scales $\sigma \in \{0, 8, 16, 32, 64, 128\}$ ms, the method tracks how event counts decay. A scaling exponent ( $\alpha_E$ ) is estimated via log-log regression to describe the persistence of structure across scales.
Mathematical Property: The transformation preserves the discrete Total Variation of the signal exactly (telescoping sum of absolute differences), ensuring that the accumulated amplitude excursion is not lost, even though intermediate samples are omitted.

Experimental Design
The evaluation utilized five PhysioNet databases to test the representation across different contexts:

AFPDB: Paired recordings of paroxysmal atrial fibrillation (PAF) to detect within-subject state changes (pre-PAF vs. PAF-onset).
NSRDB: Normal sinus rhythm for stable reference.
CHFDB: Severe heart failure for chronic pathological context.
CUDB: Ventricular tachyarrhythmia to test for common pre-fibrillation transitions.
NSTDB: Controlled electrode-motion noise added to clean signals to test if noise simply maximizes event density.

Key Results

Visual Distinctiveness: FOXTAIL ensembles revealed distinct organizational patterns across databases. For instance, CUDB (tachyarrhythmia) showed fewer retained events but larger differential magnitudes compared to NSRDB or AFPDB, demonstrating that low retention does not equate to low signal activity.
Within-Subject State Change (AFPDB): In paired PAF recordings, the transition to PAF was associated with:
- A slight decrease in the median retention ratio ( $\rho$ ).
- A significant increase in the instability of retention from beat to beat ( $SD(\Delta\rho)$ ).
- A shallower decline in event count with smoothing (lower $\alpha_E$ ), indicating greater persistence of structure across scales.
Noise vs. Complexity (NSTDB): Contrary to the hypothesis that noise simply increases turning-point density, adding electrode-motion artifact (lowering SNR) reduced the typical retention ratio while increasing between-beat dispersion and differential magnitude. This indicates that dense event patterns are not automatically equivalent to physiological complexity or noise; the nature of the disturbance matters.
Pre-VF Transition (CUDB): No common, robust short-term transition was found in the final two minutes before annotated ventricular fibrillation onset across the cohort. Individual trajectories varied significantly, suggesting that a universal "final signature" does not exist for this dataset.

Significance and Claims
The paper explicitly states that FOXTAIL is not proposed as a replacement for diagnostic ECGs or as a new classifier. Instead, its primary contribution is the establishment of a new observation and measurement domain.

Observational Strategy: It allows researchers to ask a more basic question: "How is the electrical organization of the heart changing from one beat to the next, and which of those changes persist across scale?"
Complementary View: It complements conventional morphology by making beat-to-beat organization, density, stability, and scale persistence visible and measurable.
Interpretation of Density: The study demonstrates that a dense event pattern is not automatically a sign of physiological complexity, measurement artifact, or impending disease; these must be interpreted jointly with amplitude, stability, and scale persistence.
Future Utility: While clinical classifiers may eventually be built on this domain, the current work focuses on validating the representation's ability to reveal state changes and distinguish signal characteristics from artifacts, laying the groundwork for future longitudinal and prospective validation.

Technical Summary: FoxTail: An R-Peak-Anchored Event Domain for Visualizing and Quantifying Changes in ECG Dynamics

More like this