Abstract
In volumetric CT, the through-plane axis is typically the mostdegraded,andinter-slicecontinuityiseasilylost,especiallyforthin bonystructures.Forthissuper-resolutiontask,neighbor-conditioned2.5D diffusion is a natural alternative to full 3D generation, but it fails when theconditioningslicesarethemselvesinformation-deficient.Oncethisinformationislostordistortedintheconditioningslices,thepriorhallucinatesdetailinconsistentlyacrossslices.WeproposeRECON(restorationguidedconditioning),whichseparatesdeterministicscaffoldrecoveryfrom diffusion-baseddetailrefinement.Alightweightresidualnetworkfirstrecovers low-to-mid-frequency anatomy and through-plane continuity, establishing a z-consistent scaffold for thin structures. A 2.5D diffusion model then synthesizes per-slice detail conditioned on these restored neighbors. At inference, null-space sampling enforces data consistency with the measured degraded observation. Under 4× degradation, RE- CONpreservesthin-structurefidelityandthrough-planecontinuity,and ondownstreamsegmentation,restoredvolumesyieldDicescorescloseto those achievable on the ground-truth high-resolution volumes.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SASHIMI_061.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/SASHIMI_061_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=xK2PZHFcql
BibTex
@InProceedings{ChuDon_RECON_MICCAISAT2026,
author = { Chu, Dong AND Becce, Fabio AND Büchler, Philippe},
title = { { RECON: Restoration-Guided Conditioning for Thin-Structure-Coherent CT Super-Resolution } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17258},
month = {pending},
page = {pending}
}