Abstract
Deep learning architectures for medical imaging often rely on encoders pre-trained on natural images despite the domain gap with 3D brain MRI. We present an extensive benchmark of 8 natural-image-pretrained and 10 medical-image-pretrained vision encoders on age/sex prediction and disease classification (ADNI, PPMI), finding 2D natural-image encoders perform most favorably even against models pretrained on over 100k brain MRI volumes.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLCN_2026_038.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=DW0ULsgbK4
BibTex
@InProceedings{FalPie_Representation_MICCAISAT2026,
author = { Falconnier, Pierre AND Trombetta, Robin AND Duchateau, Nicolas AND Lartizien, Carole},
title = { { Representation Learning for 3D Brain Imaging: A Benchmark } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17255},
month = {pending},
page = {pending}
}