Abstract
Pediatric glioma recurrence risk stratification at diagnosis could transform surveillance, sharpen treatment escalation, and guide follow-up planning. Beyond serial Magnetic Resonance Imaging (MRI) surveillance, routine hematoxylin and eosin (H&E) histopathology may offer complementary recurrence-relevant information at the cellular level already available at diagnosis. Pathology foundation models have shown early promise in neuropathology, particularly in generalizability, yet rare cancers such as pediatric gliomas are underrepresented in these models and lack sufficient recurrence-positive cases at any single center to tune them locally. Federated learning addresses this constraint by enabling multi-institutional collaboration without centralizing patient data. In this work, we introduce FABRIC (Federated Assessment of Brain tumor Recurrence In Children), which predicts patient-level recurrence risk by applying double-tier feature distillation, grouping patches into pseudo-bags to prioritize recurrence evidence, and federates site updates through sample-size-weighted aggregation while keeping patient data local. Across four consortium and institutional cohorts from the United ⋆ ⋆⋆ Jiying Wang and Anbang Liu contributed equally as co-first authors. Corresponding author. 2 J. Wang et al. States, Europe, and Asia, FABRIC achieved the highest mean AUC and balanced accuracy over standard foundation model-enabled analytic histology pipelines, improving discrimination by up to 2.37% in AUC and 3.11% in balanced accuracy while preserving class-balanced performance. These results establish FABRIC as a privacy-aware, histologically interpretable, and generalizable framework for pediatric glioma recurrence prediction and broader rare-disease prognosis.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/PedAItrics_007.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{WanJiy_PrivacyPreserving_MICCAISAT2026,
author = { Wang, Jiying AND Liu, Anbang AND Ji, Zheyi AND Chikontwe, Philip AND Yang, Jiancheng AND Zhu, Biyue AND Pytel, Peter AND Damodharan, Sudarshawn AND Xu, Ziyue AND Yu, Kun-Hsing AND Volchenboum, Samuel L. AND Zhao, Junhan},
title = { { Privacy-Preserving Federated Distillation of Foundation Models for Multi-Institutional Pediatric Glioma Recurrence 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}
}