Abstract
We investigated the effect of incorporating contrast-informed augmentations into pretraining data on the adult-to-neonatal adaptation behavior of deep learning-based magnetic resonance (MR) image reconstruction models. Models were pretrained on unaugmented adult data, contrast-informed augmented adult data, or a mixture of both, then fine-tuned with varying amounts of neonatal data. Evaluation on held-out neonatal and adult test sets assessed neonatal adaptation and adult-domain retention. Models pretrained on a mix of unaugmented and augmented adult data produced the best zero-shot generalization to neonatal data. Models pretrained on exclusively augmented adult data reduced adult-domain performance and did not consistently improve zero-shot neonatal performance. All pretrained models performed significantly better on neonatal reconstruction after fine-tuning with neonatal data and generally exceeded neonatal-only training in low-data settings. These results suggest that contrast-informed augmentation can improve neonatal MR image reconstruction when incorporated into a mixed pretraining distribution rather than used as a full replacement for real adult data. MR Reconstruction Domain Adaptation Infant Imaging.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/RIME_023.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=MY7yeG6Sub
BibTex
@InProceedings{MooSte_ContrastReversal_MICCAISAT2026,
author = { Moore, Stephen AND Leijser, Lara AND Frayne, Richard AND Souza, Roberto},
title = { { Contrast-Reversal Augmentation for Adult-to-Neonatal MR Image Reconstruction: Pretraining and Fine-Tuning Strategies } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17277},
month = {pending},
page = {pending}
}