Dynamical Collapse in F-Wave Vectorcardiography Signals Precedes Spontaneous Atrial Fibrillation Termination

Abstract 2305054, presented at Western Atrial Fibrillation Symposium 2026

The mechanisms of spontaneous atrial fibrillation (AF) termination remains highly debated. Competing theories include a random stochastic event or a structured, dynamical collapse. Direct electrophysiologic evidence distinguishing between spontaneous and pre-terminating states is limited. We hypothesized that human AF demonstrates measurable dynamical changes immediately before spontaneous termination detected with Principal Component Analysis (PCA)-based f-wave vectorcardiography (VCG) and machine-learning classification.We analyzed thirty long-duration ECG records from the PhysioNet AF Termination Challenge Database: 10 terminating [T], 10 one-minute pre-termination [S], 10 non-terminating [N]) from 20 subjects. After QRS removal, f-waves were projected into a 3D VCG. PCA identified optimal orthogonal axes (PC1, PC2, d(PC1)/dt) for measuring dynamical collapse. Dynamic features of the f-wave loop trajectory, including area, directional variance, turbulence, peak values, and terminal slopes, were extracted. Random forest classifiers evaluated for differences between AF states and feature importance was determined by permutation analysis.A reproducible contraction in VCG organization occurred only in the final seconds before AF termination. T vs S VCGs were highly separable (AUC: 0.85 +/- 0.25), which was driven mostly by features reflecting abrupt dynamical collapse. This includes loss of activation irregularity, wavefront propagation, and electrical volume. In contrast, T vs N and S vs N comparisons resulted in poor discriminative performance (AUC: 0.40 & 0.20), indicating that AF one minute pre-termination is electrophysiologically indistinguishable from sustained AF.Our findings suggest a pre-collapse state with abrupt declines in electrical volume and velocity precedes AF termination. PCA-based f-wave VCG features identified this global instability, revealing a brief mechanistic window before sinus restoration that may inform future AF prediction algorithms.