Abstract
The reconstruction kernel of a CT scan governs its spatial resolution and noise texture, and mismatched kernels degrade quantitativeanddownstreamanalysis.Wepresentamethodtoconvertanimage of any reconstruction kernel to a single reference kernel (any-to-one) – spanning both intra-vendor and cross-vendor kernels – while requiring only imagesreconstructedwiththatreferencekerneltobuildthemodel. Insteadofpairedmulti-kernelscansoranysource-kerneldata,wesynthesize(softinput,sharptarget)trainingpairsbyapplyingaradialspectral degradation to reference-kernel images: a frequency-domain radial lowpass that reshapes the noise power spectrum to follow real soft kernels more closely than a Gaussian blur. Concentrating the synthetic severity ontheregimeofclinicallyusedsoftkernels(hard-biasedsampling)yields asinglegeneratorthatmapsabroadrangeofinputsoftnesstothereferencekernel.OnapairedBr40f→Br60fchest-CTtestsetof238volumes, themodelreachesSSIM0.929/RMSE58.2HU,competitivewiththe strongest unpaired baseline that additionally uses source-kernel images; anablationisolatesthecontributionofeachdesignchoice.Byremoving the per-kernel data requirement, the method offers a practical route to kernel harmonization when only reference-kernel images are available.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SASHIMI_033.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=zQxqAS7AVD
BibTex
@InProceedings{JunJi_AnytoOne_MICCAISAT2026,
author = { Jung, Ji-Hoon AND Lee, Sang Min AND Lee, June-Goo},
title = { { Any-to-One CT Kernel Conversion without Source-Kernel Data } },
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}
}