Abstract
Through-plane super-resolution (SR) of brain magnetic resonance imaging (MRI) is complicated in practice by two compoundingchallenges:thefrequentunavailabilityofanauxiliaryhigh-resolution (HR) contrast at inference, and the prohibitive sampling cost inherent to diffusionand flow-based generative models. We present Dualdistillation Reference-free one-step Image Flow for MRI Through-plane SR(DRIFT),aone-stepframeworkenablingreference-freeinferencefor through-planebrainMRISRthroughaunifieddualdistillationstrategy. A rectified flow teacher is first trained to map noise-augmented lowresolution (LR) inputs to HR outputs, conditioned on both the LR inputandtheauxiliaryHRcontrast,overmultipleODEsteps.Aone-step student is then distilled from the teacher via ODE trajectory alignment with stochastic reference dropout: during distillation, the auxiliary contrastinputisrandomlyreplacedbyanullcondition,forcingthestudent tointernalizethestructuralknowledgethattheteacherextractsfromthe auxiliary contrast without requiring it at inference. This draws a principled analogy to classifier-free guidance and, to our knowledge, is the first flow-based approach to simultaneously distill away both auxiliary conditioninginputsandsamplingstepsformedicalimageSR.Extensive experiments demonstrate that our method consistently outperforms existingMRISRmethodsinreconstructionqualityandinferenceefficiency.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SASHIMI_004.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=P2mGS2NaQt
BibTex
@InProceedings{ZhaZhe_DRIFT_MICCAISAT2026,
author = { Zhang, Zheng AND Zhou, Zechen},
title = { { DRIFT: Dual-Distillation Reference-Free One-Step Image Flow for Brain MRI Through-Plane 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}
}