Abstract
Recognizing heterogeneous histomorphological regions within glioblastoma (GBM) at the patch level remains difficult because of class imbalance, variation between patients, and similarities in local tissue morphology. We evaluate whether a frozen pathology foundation model could provide useful representations for classifying ten GBM tissue categories. Each 512 × 512-pixel haematoxylin-and-eosin-stained patch is encoded by the ALICE vision stage as a 3,840-dimensional feature vector. These features are used to train a lightweight classifier with a SiLU activation and a linear output layer. Hierarchical class–patient–slide balanced sampling is applied during training. We conduct patient-disjoint five-fold cross-validation using 1,402,607 patches from 244 slides of 121 patients. The trained model is then evaluated on the official online validation set. In the online validation set, our model achieves F1 score 0.757 and MCC 0.896. Performance differs considerably between tissue categories. Some categories are recognized reliably, while others remain challenging. Overall, the results show that frozen ALICE vision-stage features can be combined with hierarchical balanced sampling for patch-level classification of GBM subregions. They also indicate that rare categories and regions with similar local morphology require further investigation. The source code is publicly available at https://github.com/zqf1234588/Brats2026_Pathology.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/BraTS_Path_004.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=O46q48ZNRx
BibTex
@InProceedings{ZhoQif_PatchLevel_MICCAISAT2026,
author = { Zhong, Qifeng AND Weng, Ying AND Chen, Ke},
title = { { Patch-Level Glioblastoma Subregion Classification with Foundation Model and Hierarchical Balanced Sampling } },
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}
}