Abstract
Glioblastoma is the most aggressive primary brain tumor, characterized by pronounced histologic and molecular heterogeneity that complicates consistent diagnosis. Recognizing patterns such as necrosis, microvascular proliferation and immune-rich areas is important for reproducible pathologic assessment. We propose the Glioma Adaptive Distillation Network (GLAD-Net), a framework for fine-grained patch-level classification of glioma tissue subtypes in H&E-stained whole slide images. GLAD-Net combines complementary knowledge from H-Optimus-0, Virchow2, Prov-GigaPath and UNI2 through adaptive teacher selection, feature alignment and gated fusion. We evaluate it on the BraTS-Path 2026 Challenge dataset, which contains 1,631,432 labeled training patches in ten classes. Model development uses patient-grouped five-fold validation based on the organizer-provided ID–Patch mapping. On the official validation set, the final ensemble achieves 0.888 accuracy, 0.931 AUROC, 0.844 MCC, 0.573 F1, 0.558 recall and 0.982 specificity. The improvement over the H-Optimus-0 baseline is modest in F1 but consistent in MCC; external institutional validation is still required before drawing clinical conclusions.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_Path_014.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=UNT0Ugbe9U
BibTex
@InProceedings{LuMen_GLADNet_MICCAISAT2026,
author = { Lu, Mengkang AND Zeng, Qingjie AND Lu, Zilin AND Xia, Yong},
title = { { GLAD-Net: Adaptive Multi-Teacher Distillation of Vision Foundation Models for Fine-Grained Glioma Histopathology } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17252},
month = {pending},
page = {pending}
}