Abstract
Multiparametric MRI holds promise for glioma grade classification. However, the relative performance of modality-defined feature sets may depend on the grade boundary evaluated and whether testing occurs at an unseen institution. We evaluated WHO grade 2/3/4 classification as a multicentre benchmark for MRI histogram features under leave-one-institution-out (LOIO) external validation. The cohort comprised 414 glioma cases across six cohorts with histopathological grade labels. Five predefined feature sets were assessed: perfusion (FS1), structural MRI (FS2), structural plus perfusion (FS3), structural plus perfusion plus ADC (FS4), and FS4 plus demographic covariates (FS5). In each LOIO fold, model development, hyperparameter tuning and model-family selection were performed only within the training institutions using a restricted core set of Elastic Net, linear support-vector machine and Random Forest classifiers. The held-out institution was used once for final evaluation. Median external macro-AUCs were closely clustered across feature sets (FS1 0.836, FS2 0.798, FS3 0.818, FS4 0.824 and FS5 0.830). Internal-to-external decay was negative for all feature sets. Class-wise analyses showed small descriptive differences: perfusion-only features were directionally favoured across sites for grade 2, multimodal feature sets were numerically higher for grade 3, and structural MRI alone matched the best grade 4 performance. Given five evaluable cohorts, these patterns are hypothesis-generating rather than evidence of feature-set superiority. Class-wise external reporting can nevertheless complement aggregate multiclass performance.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MLMI_067.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=jp31uNCHAa
BibTex
@InProceedings{SiaLoi_Beyond_MICCAISAT2026,
author = { Siakallis, Loizos AND Bisdas, Sotirios},
title = { { Beyond Global Macro-AUC: Class-Wise Modality Contribution in Multicentre MRI-Based Glioma Grading } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17272},
month = {pending},
page = {pending}
}