Abstract
Cortical parcellation provides a systematic way of dividing the cortex into meaningful regions for region-wise analysis. Although automatic parcellation is promising, recent learning-based approaches depend on large-scale datasets, which requires expensive data acquisition. To tackle this challenge, various data augmentation methods have been explored; nevertheless, existing approaches suffer from two key limitations: (1) insufficient diversity in the augmented patterns and (2) the lack of corresponding labels for synthesized samples. In this work, we propose a DDPM-based augmentation framework called Spherical RePaint that synthesizes diverse cortical surface features while directly reusing known parcellation labels. Specifically, our method integrates a label-derived boundary mask with spherical inpainting, preserving real features near parcel boundaries to maintain label validity while generating diverse patterns within region interiors. During sampling, an iterative resampling procedure re-applies forward diffusion to better harmonize preserved boundaries with synthesized interiors. Experimental results demonstrate that our approach achieves the highest Dice scores for cortical parcellation across all evaluated few-shot settings under extreme data scarcity.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ShapeMI_010.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=XEFUyI5K2A
BibTex
@InProceedings{KimKan_Spherical_MICCAISAT2026,
author = { Kim, Kangmin AND Kim, Jongmin AND Son, Jiwon AND Choi, Junho AND Lyu, Ilwoo},
title = { { Spherical RePaint: Label-Reusable Diffusion-Based Augmentation for Parcellation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17259},
month = {pending},
page = {pending}
}