Abstract
The ASNR-MICCAI BraTS Inpainting Challenge aims to
synthesize pseudo-healthy brain tissue within masked regions of T1-
weighted MRI, facilitating downstream neuroimaging applications that
assume healthy anatomy. To tackle this challenge, we propose GACWaveDiff,
a geometry-aware conditional 3D wavelet diffusion framework.
Built upon fastWDM3D, GAC-WaveDiff operates in the 3D Haar-wavelet
domain and retains its efficient two-step reverse process. To reduce spatial
ambiguity during patch-based reconstruction, we condition the network
on normalized global 3D coordinates that encode the anatomical
location of each input window. We further incorporate a signed
distance transform to provide continuous information about missingregion
geometry and boundary proximity. Both conditions are spatially
aligned with the Haar-wavelet coefficient grid. To improve generalization
across corruption patterns, we introduce a mask- and anatomy-aware
augmentation strategy that varies mask location, orientation, morphology,
and topology, including multifocal configurations. During inference,
intermediate-noise-suppressed sampling is combined with flip-based testtime
augmentation to improve voxel-wise reconstruction fidelity. On the
BraTS 2026 online validation leaderboard, GAC-WaveDiff achieves a
PSNR of 23.38 dB, improving the PSNR by 3.68 dB over the reproduced
fastWDM3D baseline. Ablation experiments show complementary
gains from geometry-aware conditioning, task-specific augmentation,
and fidelity-oriented inference. These results demonstrate the value
of explicitly modeling anatomical position and missing-region geometry
in efficient volumetric MRI inpainting. Our code can be accessed on:
https://github.com/EvelyneCalista/GAC-WaveDiff.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_Inpainting_011.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/BraTS_Inpainting_011_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=vq2rTWLAd6
BibTex
@InProceedings{CalEve_GeometryAware_MICCAISAT2026,
author = { Calista, Evelyne AND Chen, Yong-Sheng},
title = { { Geometry-Aware Conditional 3D Wavelet Diffusion for Pseudo-Healthy Brain MRI Inpainting } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17254},
month = {pending},
page = {pending}
}