Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
March 14, 2026npj Precision OncologyOpen Access

Multimodal fusion of pathology and radiology foundation models for WHO 2021 glioma subtyping

View Full Paper
Ask AI
Bookmark
Share

Authors

CSCamillo SaueressigDSDaniel ScholzPRPhilipp Raffler

Discussion

Loading...

Member takes

Overview

Demonstrates improved glioma subtyping in unseen cases by integrating pathology and MRI data, highlighting the potential for effective multimodal classification.

Key Points

  • The research aims to enhance glioma subtyping accuracy using a multimodal classification framework that merges histology and MRI data.
  • Developed a classification framework using histopathology and MRI
  • Trained on 772 histopathology cases and 959 MRI scans
  • Evaluated three modality fusion strategies with a mixture-of-experts architecture
  • Tested on 171 unseen patient-matched cases
  • Multimodal models outperformed unimodal counterparts
  • Achieved AUC scores of 0.98 during validation and 0.94 in an independent test set
  • High-performing models were effective even without paired multimodal data
  • Identified distinct visual biomarkers linked to glioma molecular subtypes

Cite This Study

Saueressig et al. (2026) studied this question.

synapsesocial.com/papers/69b4fb9db39f7826a300bf9bhttps://doi.org/10.1038/s41698-026-01366-5
View Full Paper
Ask AI
Bookmark
Share