Abstract
Multimodal pathology combines histology with molecular assays,
clinical variables, and free-text reports. In deployment, however,
modality availability is heterogeneous: expensive or delayed tests are
often missing, yet many multimodal approaches assume a fixed input
panel at inference. Moreover, reported multimodal gains are frequently
attributed to cross-modal synergy without quantifying interaction effects
under a controlled protocol. We present FIRB, a Fusion, Interaction
and Robustness Benchmark for multimodal learning in computational
pathology. To this end, we introduce an any-subset benchmarking framework
that standardizes modality preprocessing and frozen encoders, exposes
a shared token interface for plug-in fusion operators, and evaluates
models under clinically motivated missing-modality regimes, including
cheap-panel inference. Within this framework, we implement nine representative
fusion paradigms spanning early, intermediate, late, and sequential
fusion, and benchmark them on four multimodal cohorts spanning
several modalities. We probe whether interaction-heavy behavior is
linked to improved performance, we compute modality-level uni-modal
and pairwise interaction rates and test their association with performance
across regimes. GitHub: https://github.com/agentdr1/FIRB
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COMPAYL_009.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
BibTex
@InProceedings{ReiDan_FIRB_MICCAISAT2026,
author = { Reisenbüchler, Daniel AND Richter, Charlotte AND Bozorgpour, Afshin AND Kumari, Pratibha AND Deng, Ruining AND Merhof, Dorit},
title = { { FIRB: Fusion, Interaction and Robustness Benchmark for Multimodal Learning in Computational Pathology } },
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}
}