Contrastive multimodal deep learning for survival prediction in grade 2/3 gliomas
Peiying Hua1, Chun-Chieh Lin2, Travis Fenlon2
1Department of Biomedical Data Science, Geisel School of Medicine at Dartmouth, Hanover, NH, United States.
JNCI Cancer Spectrum
|April 16, 2026
Summary
This study introduces a multimodal deep learning model for improved survival prediction in grade 2/3 glioma patients. The novel approach integrates histopathology, genomics, and clinical data, outperforming existing methods for better patient stratification.
Area of Science:
- Neuro-oncology
- Computational Biology
- Medical Imaging
Background:
- Accurate survival prediction for grade 2/3 glioma is challenging due to tumor heterogeneity.
- Current prognostic methods using single-modality data have limitations.
Purpose of the Study:
- To develop a multimodal deep learning framework for enhanced survival prediction in glioma.
- To integrate histopathology images, somatic mutations, and clinical data for improved prognostic accuracy.
Main Methods:
- A deep learning framework was developed integrating whole-slide histopathology images, somatic mutations, and clinical-demographic data.
- A three-stage training pipeline combined contrastive learning with survival-specific optimization.
- The model was trained on 498 TCGA patients and validated on an independent DHMC cohort.
Main Results:
- The contrastive multimodal model achieved a c-index of 0.91, significantly outperforming unimodal models (0.76-0.87).
- Kaplan-Meier analysis showed clear survival separation across risk strata (P=4.4×10-5).
- External validation on the DHMC cohort yielded a c-index of 0.87 after domain adaptation.
Conclusions:
- Contrastive multimodal learning significantly improves survival prediction in grade 2/3 gliomas.
- This annotation-free approach facilitates early risk stratification and personalized treatment decisions.
- The framework shows promise for clinical trial stratification using routinely collected data.
