PhD Scientific Days 2026

Cardiovascular Medicine and Research 1.

Machine-learning analysis of paced P-wave morphologies to identify atrial ectopy origin in atrial fibrillation patients

Előadó neve

Dr. Komlosi, Ferenc

Neptune code

FK65F3

Előadó munkahelye

Semmelweis University Heart and Vascular Center

Előadó telefonszáma

+36308962486

Előadó e-mail címe

gyuszkob@gmail.com

Az előadás címe

Machine-learning analysis of paced P-wave morphologies to identify atrial ectopy origin in atrial fibrillation patients

Szerző(k) neve és munkahelye

Gyula Bohus Dr.¹, Ferenc Komlósi Dr.¹, Imre Szakal Dr.¹, Alexandra Karsai¹, Bence Arnóth Dr.¹, Helga Sánta¹, Áron Csenák¹, István Osztheimer Dr. PhD¹, Nándor Szegedi Dr. PhD¹, Péter Perge Dr. PhD¹, Zoltán Salló Dr. PhD¹, Edit Tanai Dr.¹, Béla Merkely Prof. Dr.¹, László Gellér Prof. Dr.¹, Klaudia Vivien Nagy Dr.¹

1: Semmelweis University

Bemutatás módja

Szóbeli

Szekció

Cardiovascular Medicine and Research 1.

Language of the presentation

English

Preferred session

Cardiovascular Medicine and Research

Összefoglaló szövege

Introduction: Atrial ectopic beats are key triggers of atrial fibrillation (AF). Pulmonary vein (PV) origin is the most common and the main ablation target. However, non-PV foci can also initiate and sustain arrhythmia. In patients with recurrent AF after PV isolation, the trigger is often uncertain, complicating ablation strategy. Thus, identifying the dominant ectopic source from surface ECG may guide ablation. We hypothesised that P-wave morphology during intracardiac pacing can train a deep learning (DL) model to identify ectopy origin.
Aims: To develop and validate a DL model capable of predicting atrial pacing sites from surface ECG P-waves.
Methods: We prospectively analysed patients undergoing first-time AF ablation. During the procedure, 30 low-rate pacing pulses were delivered from eight predefined atrial sites (four PVs, superior vena cava, coronary sinus ostium, left atrial posterior wall, and appendage) with simultaneous 12-lead ECG recording. The P-wave (200 ms from stimulus) was analysed in leads I, II, III, aVF, and V1. Noise filtering used discrete wavelet transform. A convolutional neural network (CNN) was trained with 5-fold cross-validation and tested using patient-wise leave-one-out validation. Performance metrics included accuracy, F1 score, and AUROC.
Results: Data from 78 patients (17,621 P-waves) across 8 sites were analysed. The CNN showed strong discrimination, with mean per-class AUROC 0.88 (0.83–0.97). This indicates that paced P-wave morphology encodes spatial origin. The average F1 score was 0.58 and accuracy 0.56. Calibration was favourable (expected calibration error 0.19; Brier score 0.64).
Conclusion: A CNN trained on paced P-waves showed excellent performance with good discrimination and calibration. This approach may enable non-invasive localisation of ectopic foci, potentially guiding targeted ablation and improving repeat procedures. Further validation in spontaneous ectopy and multicentre settings is needed.

University

Semmelweis University

Supervisor

Dr. Klaudia Vivien Nagy

Publication of my abstract

I do not give consent to the publication of my abstract on the website of the congress.

phd.section.field

in doctoral studies after complex exam (PhD)

Kind

Szabad

Status

elfogadva

Accepted presentation method

szóbeli

Előadás fájl jóváhagyás

nem rendelkezett róla

Előadó

9838

Start

16:30

End

16:40