Abstract
Automated glioma segmentation models trained on high-quality MRI data generalise poorly to African clinical settings, where 1.5T scanners and motion artefacts are prevalent. We investigated cross-modal leave-one-out (LOO) consistency as a label-free reliability signal for SegResNet glioma segmentation on BraTS-Africa (95 cases; 76 training / 19 validation). Standard LOO consistency was strongly confounded by FLAIR architectural dependence: FLAIR removal produced the largest disruption in 15/19 cases (79%) regardless of segmentation quality, yielding a non-significant correlation with ground-truth Dice (r = 0.240, p = 0.322). Excluding FLAIR and computing T1-family consistency yielded a higher Pearson correlation (r = 0.660, p = 0.002), but this was not robust under Spearman analysis (ρ = 0.202, p = 0.408), bootstrap CIs spanned zero, and Steiger’s test was non-significant (p = 0.100). Binary triage classifiers performed below chance under leave-one-patient-out cross-validation (AUC = 0.350). These findings identify FLAIR dependence as a critical confound in cross-modal consistency frameworks and motivate modality-aware reliability estimation for safe AI deployment in low-resource neuro-oncology.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_037.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MIRASOL_037_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=VAuLfVkJNF
BibTex
@InProceedings{MfeOfi_FLAIR_MICCAISAT2026,
author = { Mfetane, Ofile Seneo AND Sepora, Tshegofatso Olorato AND Moerane, Aobakwe AND Kayuna, Matthew AND Oats, Unotjari AND Lungowe, Sililo AND Makgasane, Magnus Tiiso AND Sebina, Seipone Talama AND Manyanda, Letso Jessica AND Chibuta, Peter AND Hassan, Maryam Olaitan AND Anazodo, Udunna C. AND Iorumbur, Aondona Moses AND Raymond, Confidence AND Sidume, Freedmore},
title = { { FLAIR Dependence Confounds Cross-Modal Consistency for SegResNet Reliability Estimation on BRaTS-Africa } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17264},
month = {pending},
page = {pending}
}