Abstract
High b-value diffusion-weighted imaging (DWI) of the breast provides high sensitivity for cancer detection, but maximum-intensity-projection (MIP) images are frequently affected by artifacts. This study assessed whether high b-value DWI MIP artifacts can be forecast from preceding low b-value DWI or T2-weighted fat-saturated (T2w-FS) acquisitions. In total, 1,089 examinations from two 3.0 T systems and 195 examinations from the multi-vendor ACRIN-6698 dataset were retrospectively analyzed. Artifact severity was rated on a five-point scale by three readers; moderate-to-severe artifacts (score ≥3) occurred in 62% of examinations. A 3D DenseNet201 model was trained on internal data and evaluated in independent internal and external test datasets. Operating thresholds were selected in internal validation data to target high specificity. In the internal test dataset, forecasting from low b-value DWI and T2w-FS achieved AUROCs of 0.75 (CI: 0.68–0.82) and 0.74 (CI: 0.67–0.81), respectively, without a significant difference between methods (p=0.55). At the fixed thresholds, sensitivity/specificity were 0.38/0.93 for low b-value DWI and 0.49/0.87 for T2w-FS. In the external dataset, AUROCs were 0.63 (CI: 0.54–0.70) and 0.71 (CI: 0.64–0.78), respectively, with better discrimination for T2w-FS (p=0.013). Sensitivity/specificity were 0.21/0.89 for low b-value DWI and 0.61/0.68 for T2w-FS. Preceding breast MRI acquisitions contain information associated with subsequent high b-value DWI MIP artifacts. However, variable external performance indicates limited transferability and the need for multi-vendor training, recalibration, and prospective evaluation before clinical use.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/Deep_Brea3th_017.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=kA3FQerhpy
BibTex
@InProceedings{LieAnd_Forecasting_MICCAISAT2026,
author = { Liebert, Andrzej AND Hadler, Dominique AND Skwierawska, Dominika AND Kapsner, Lorenz A. AND Folle, Lukas AND Das, Badhan AND Schreiter, Hannes AND Ohlmeyer, Sabine AND Maier, Andreas AND Laun, Frederik B. AND Uder, Michael AND Wenkel, Evelyn AND Bickelhaupt, Sebastian},
title = { { Forecasting Artifacts in High b-Value DWI MIP Images of the Breast Using Deep Learning and T2-Weighted or Low b-Value Acquisitions } },
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}
}