Abstract
Marfan syndrome (MFS) is a rare connective-tissue disorder whose diagnosis can be challenging when systemic manifestations are mild. Although craniofacial morphology contributes to the phenotype, facial characteristics are often subtle and spatially distributed. We propose a conditional mesh variational autoencoder (cVAE) framework that unifies three-dimensional facial screening and anatomical explanation. An age- and sex-conditioned representation model learns latent embeddings of registered facial meshes for disease screening through a linear classifier and decoder-based counterfactual explanations for result interpretability. A second instance of the same architecture, additionally conditioned on disease status, generates synthetic facial meshes. These synthetic data are evaluated as an alternative training source through train-on-synthetic, test-on-real experiments, providing a quantitative assessment of generative fidelity. Experiments on 1548 registered facial meshes (1423 controls and 125 individuals with MFS) demonstrate accurate screening, anatomically meaningful explanations, and high agreement between real- and synthetic-trained models in terms of latent organization, facial phenotype, and decision-boundary counterfactuals. These results show that a single conditional generative framework can simultaneously support explainable disease screening and quantitative validation of synthetic-data fidelity for rare-disease applications. Code and the trained screening model are available at https://github.com/gfacchi-dev/MICCAI-MFS-MeshVAE
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ShapeMI_027.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=GYOxXA0RzK
BibTex
@InProceedings{FacGiu_Learning_MICCAISAT2026,
author = { Facchi, Giuseppe Maurizio AND Lanzarotti, Raffaella AND Grossi, Giuliano AND Pini, Alessandro AND Dolci, Claudia},
title = { { Learning the Marfan Face: A Conditional Mesh VAE for Explainable Screening and Synthetic Generation } },
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}
}
