Abstract
Cardiac magnetic resonance analysis requires consistent reasoning across anatomy, cine dynamics, scar, and clinical endpoints. We present CardioWorld-GQ, an observation-conditioned cardiac imaging world model whose differentiable readout computes volumes, mass, scar burden, and all-frame LVEF from the same probability field as the segmentation. On the official validation split, the full model attains mean Dice $0.780$, mean IoU $0.661$, ASSD $2.49$ mm, soft LVEF MAE $0.47\%$, and $91.2\%$ marginal coverage at a nominal $90\%$ target. Ablations show metric-dependent trade-offs rather than uniform superiority. The framework targets L1-L2 imaging functionality with a limited L3-like differentiable query, without claiming treatment simulation or control.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CMRSeg_040.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/CMRSeg_040_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=htyvjgiNJ4
BibTex
@InProceedings{YuaJin_CardioWorldGQ_MICCAISAT2026,
author = { Yuan, Jing AND Wang, Shengming AND Liu, Ningtao},
title = { { CardioWorld-GQ: An Observation-Conditioned Cardiac Imaging World Model with a Differentiable Clinical Readout and Calibrated Endpoint Uncertainty } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17268},
month = {pending},
page = {pending}
}