Abstract
Integrating Whole Slide Images, radiological scans (CT and
MRI), genomic profiles, and clinical metadata offers significant potential
for outcome prediction in oncology. However, real-world multimodal
datasets are characterized by extreme modality missingness, heterogeneous
acquisition protocols, and high computational demands.
We propose a structurally consistent multimodal framework for survival
prediction that natively operates under incomplete modality configurations
without reconstructing missing modalities. Each modality is independently
encoded and projected into a shared latent patient space,
where representations are fused via element-wise summation or concatenation.
This design avoids the complexity of reconstruction-based strategies
while effectively mitigating representation shift.We evaluate the proposed
framework on the MMIST-ccRCC dataset, reflecting realistic scenarios
where modality missingness scales from 25% for genomic profiles
to more than 90% for MRI, posing substantial challenges for multimodal
learning.
For the task of survival prediction, our model achieves higher Balanced
Accuracy compared to the dataset’s original baseline across all evaluated
modality subsets. Without requiring the reconstruction of missing
data, our approach provides a more computationally efficient solution,
reducing training time while maintaining robust prognostic performance
across all modality combinations. These results demonstrate that enforcing
structural consistency in a shared latent space offers a computationally
efficient alternative for multimodal outcome prediction in clinically
realistic settings with severe missing data. The source code is publicly
available at https://github.com/AI-BioInformatics/OXA-MISS.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COMPAYL_050.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{MicFra_Unified_MICCAISAT2026,
author = { Miccolis, Francesca AND Corso, Giulia AND Pipoli, Vittorio AND Ficarra, Elisa AND Lovino, Marta},
title = { { Unified Multimodal Fusion for Cancer Survival Prediction: A Missing-Robust Framework across Histopathology, Radiology, Genomics, and Clinical Data } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17251},
month = {pending},
page = {pending}
}