Abstract
Deep learning algorithms for fibro-glandular tissue (FGT) segmentation in breast MRI achieve good performance, however results can vary depending on the training data and preprocessing strategies. This study assesses the generalizability of deep learning architectures in FGT segmentation on a multi-centric, multi-vendor dataset with respect to breast denisty class and preprocessing steps, which included intensity clipping, contrast adjustment, histogram matching and breast masking. Our results show that architectures’ responses varied across preprocessing steps. Among the tested models Swin UNETR achieved the highest weighted DICE scores among the evaluted networks - 0.80, while nnUNet achieved the lowest Hausdorff distance - 37.51. Among the tested configurations, breast masking and intensity clipping generally improved DICE scores of the models.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/Deep_Brea3th_033.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=4VnHB3ahi0
BibTex
@InProceedings{RydGrz_Impact_MICCAISAT2026,
author = { Rydzyński, Grzegorz AND Markale, Ameya AND Heidarikahkesh, Shirin AND Ehring, Chris AND Döppmann, Lorenz AND Hadler, Dominique AND Kapsner, Lorenz A. AND Ohlmeyer, Sabine AND Uder, Michael AND Jeleń, Łukasz AND Ivanovska, Tatyana AND Bickelhaupt, Sebastian AND Liebert, Andrzej},
title = { { Impact of preprocessing strategies on segmentation of fibroglandular tissue on breast MRI using deep learning } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17256},
month = {pending},
page = {pending}
}