Abstract
Prenatal brain development involves rapid cortical growth and folding, and disruptions in sulcal patterns have been linked to neurodevelopmental dis-orders and long-term functional impairments. The spatial organization of early sulcal folds remains relatively stable over development, making sul-cal patterns a promising substrate for detecting developmental deviations. In this work, we propose SynthSPA, an unsupervised framework for detect-ing sulcal abnormalities based on normal data. SynthSPA leverages latent diffusion-based generative modeling to synthesize diverse pseudo-abnormal (PA) sulcal pattern graphs through controlled perturbations of normal graphs and learns to discriminate them from normative samples. The synthesized PA graphs are perturbed while remaining close to the true nor-mative graphs, preserving the overall organization while introducing subtle deviations. This enables SynthSPA to learn fine-grained decision bounda-ries that are sensitive to mild and heterogeneous sulcal abnormalities. Fur-thermore, to capture the unique geometric and topological properties of sulcal patterns, SynthSPA incorporates a newly proposed heterophily-aware variational graph autoencoder (HA-VGAE), tailored to low-homophily graph structures. We train SynthSPA on publicly available healthy brain magnetic resonance imaging datasets (total n=1,542), and evaluate it on congenital heart disease (CHD), the most common birth defect associated with altered brain development. Experimental results demonstrate that SynthSPA effectively captures subtle and heterogeneous deviations and outperforms existing baselines across hemispheres and CHD subtypes. To our knowledge, this work represents the first application of diffusion-based anomaly detection to sulcal pattern analysis. Code is available at: https://github.com/hookhy/SynthSPA
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/GRAIL_006.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=tmYIIZFYsu
BibTex
@InProceedings{KwoHye_Diffusionbased_MICCAISAT2026,
author = { Kwon, Hyeokjin AND Morton, Sarah U. AND Newburger, Jane W. AND Lee, Jong-Min AND Grant, P. Ellen AND Im, Kiho},
title = { { Diffusion-based Perturbation Modeling for Unsupervised Detection of Sulcal Pattern Alterations } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17278},
month = {pending},
page = {pending}
}