Abstract
Automated segmentation enables large-scale curation of retrospective imaging cohorts, but unrecognized failures can bias downstream analyses, particularly in pediatric neuro-oncology, where multiinstitutional datasets are small and longitudinal tumor burden drives response assessments. We study quality control (QC) for automated diffuse midline glioma segmentation, benchmarking methods to identify poor segmentations. We compare supervised QC predictors based on a combination of acquisition metadata, radiomics, MRI volumes with predicted masks, and foundation-model MRI embeddings, Dice estimation, and simple model-derived baselines using inter-model disagreement, softmax entropy, and predicted tumor volume. Crucially, we assess QC by its effect on a clinical endpoint: radiotherapy-response labels derived from longitudinal tumor-volume change. Dice estimation is the strongest failure detector by AUPRC (0.933), but volume-based thresholding recovers more label agreement per re-annotation. At the endpoint level, uncorrected segmentations systematically over-assign favorable response categories relative to manual references, and volume-based selective re-annotation of scans recovers agreement at low review burden. These findings argue for endpoint-aware segmentation QC, optimized for the clinical labels segmentations support rather than for Dice prediction alone. 2 D. Laslo et al.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/PedAItrics_024.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{LasDar_Enhancing_MICCAISAT2026,
author = { Laslo, Daria AND Strijbis, Victor IJ AND Haddadi Avval, Atlas AND Vogt, Franziska AND Fathi Kazerooni, Anahita AND Jiang, Zhifan AND Parida, Abhijeet AND Kann, Benjamin AND Linguraru, Marius George AND Franson, Andrea AND Müller, Sabine AND Rauschecker, Andreas M. AND Jutzeler, Catherine R. AND Brüningk, Sarah},
title = { { Enhancing Clinically-relevant Quality Control Through Automated Segmentation Performance Prediction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17257},
month = {pending},
page = {pending}
}