Abstract
High-risk non-muscle-invasive bladder cancer (HR-NMIBC) is associated
with frequent progression after Bacillus Calmette-Guérin (BCG) immunotherapy,
making accurate risk prediction important for treatment planning and
surveillance. Current clinical diagnosis relies on pathological phenotype for staging
while molecular genotype plays an important role in BCG response. Multimodal
learning could therefore improve prediction by integrating histopathology,
RNA-seq, and clinical data, but it remains unclear which fusion strategy is most
effective in limited-size clinical cohorts. In addition, RNA-seq is often unavailable
in clinical practice; designing architectures that remain robust to missing-modality
conditions is crucial. In this study, we systematically compared multimodal
fusion strategies for time-to-event prediction in HR-NMIBC using 272 patients
with complete histopathology whole-slide images, bulk RNA-seq, and clinical
data from the CHIMERA challenge. We evaluated unimodal, bimodal, and trimodal
Cox survival models using concatenation, gated fusion, low-rank bilinear
fusion, and cross-attention. Multimodal fusion improved performance in most
configurations compared with unimodal models, with the best performance
achieved by trimodal gated fusion, with a C-index of 0.725 ± 0.083. We further
found that modality representation affected performance and that RNA dropout
improved robustness when RNA-seq was unavailable at inference time. These
findings highlight the importance of fusion design for clinically realistic multimodal
survival prediction.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COMPAYL_074.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/COMPAYL_074_supp.pdf
Link to Open Review
BibTex
@InProceedings{HasSae_When_MICCAISAT2026,
author = { Hashemi, Saeedeh AND Zuiverloon, Tahlita C. M. AND Khalili, Nadieh},
title = { { When RNA is Missing: A Multimodal Fusion Benchmark for BCG Progression Prediction in HR-NMIBC } },
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}
}