Abstract
Atrial fibrillation (AF) in the intensive care unit (ICU) is associated with stroke, heart failure, and increased mortality. Although continuous ECG monitoring is routine, this data is not used to predict AF, which could enable preventive treatment. We propose a deep learning framework that combines a Residual Network (ResNet) with Mamba, a state-space model for long-range sequence modelling, to predict AF within a two-hour horizon from raw single-lead ICU ECG. We compare the ResNet–Mamba hybrid with a ResNet-only baseline using 10- and 30- minute input windows on two ICU cohorts (MIMIC-III and Amsterdam UMC). With 30-minute windows, ResNet–Mamba reached a ROC-AUC of 0.846 and an average precision of 0.519, outperforming the ResNetonly model (0.830 and 0.446). These results suggest that selective longrange modelling of raw single-lead ICU ECG is a promising approach for early AF warning within a clinically actionable horizon.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AMAI_060.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/profile?id=%7EMaria_Galanty1
BibTex
@InProceedings{GalMar_Prediction_MICCAISAT2026,
author = { Galanty, Maria AND van der Ster, Björn AND Reuland, Merijn C. AND Hulleman, Michiel AND Vlaar, Alexander P.J. AND Sánchez, Clara I.},
title = { { Prediction of Atrial Fibrillation Onset from Continuous Intensive Care Unit ECG Monitoring: A Deep Learning Approach } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17273},
month = {pending},
page = {pending}
}