Abstract
Predicting long-term cardiac rhythm outcomes from baseline imaging remains a major challenge in atrial fibrillation (AF) management. In this study, we propose a neural network architecture, “AF-Echo”, based on spatio-temporal convolution and video self-attention, to predict cardiac rhythm status two years after initial diagnosis using transthoracic echocardiography (TTE) videos. We analyze a cohort called FASTRHAC which consists of patients hospitalized for AF and followed over a two-year period. The binary ground-truth label was defined by experts at the 24-month follow-up (M24), distinguishing sinus rhythm (healthy) from AF rhythm (non-healthy). The model is trained in a multi-task learning setting, jointly predicting rhythm outcome and auxiliary cardiological parameters. Our quantitative results show that echocardiography allows long-term rhythm prediction, yielding significantly better predictive performance than a baseline ECG-based model, highlighting the potential of deep video-based models for early risk stratification in AF management.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_050.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/forum?id=tXFBb8hYRW
BibTex
@InProceedings{SadKim_LongTerm_MICCAISAT2026,
author = { Sadreddini, Kimia AND Newson, Alasdair AND Soulat-Dufour, Laurie AND Ponnaiah, Maharajah AND Cohen, Ariel AND Hatem, Stephan},
title = { { Long-Term Prediction of Atrial Fibrillation Severity from Echocardiography Video } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17272},
month = {pending},
page = {pending}
}