Dynamic Graph and Multi-Stage Mamba Fusion for Multimodal Survival Prediction

Multimodal survival prediction using whole-slide images (WSIs) and genomic profiles is important for cancer prognosis; however, integrating gigapixel histopathology with high-dimensional molecular data remains challenging. Existing methods often rely on dense or fixed fusion strategies that may introduce redundant interactions and inadequately capture patient-specific pathology--molecular associations. We propose a sample-adaptive WSI-omics framework that constructs sparse pathway--patch interactions and combines graph-based refinement with selective state-space aggregation for patient-level survival prediction. The framework was evaluated on five TCGA cohorts using patient-level 5-fold cross-validation. Compared with SurvPath, the proposed method improved the average C-index from 0.609 to 0.638, corresponding to an absolute gain of 0.029. The improvement reached significance only in BLCA under the paired \((t)\)-test, indicating cohort-dependent statistical support. Matched profiling showed lower memory consumption than the Transformer replacement, while case-level visualizations provided model-derived pathway--patch associations. These results support sample-adaptive sparse interaction modeling for WSI-omics survival prediction. External validation, repeated-seed evaluation, and pathologist-guided interpretation remain necessary before clinical application.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-18
DOI
https://doi.org/10.21203/rs.3.rs-11124211/v1
Primary Topic
AI in cancer detection
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Dynamic Graph and Multi-Stage Mamba Fusion for Multimodal Survival Prediction

Yun Gu, Wei Xu, Jie Yang, Rui Wang
3 links
Research Square
AI in cancer detection
preprint

Dynamic Graph and Multi-Stage Mamba Fusion for Multimodal Survival Prediction

Yun Gu, Wei Xu, Jie Yang, Rui Wang
preprint en

Abstract

Multimodal survival prediction using whole-slide images (WSIs) and genomic profiles is important for cancer prognosis; however, integrating gigapixel histopathology with high-dimensional molecular data remains challenging. Existing methods often rely on dense or fixed fusion strategies that may introduce redundant interactions and inadequately capture patient-specific pathology--molecular associations. We propose a sample-adaptive WSI-omics framework that constructs sparse pathway--patch interactions and combines graph-based refinement with selective state-space aggregation for patient-level survival prediction. The framework was evaluated on five TCGA cohorts using patient-level 5-fold cross-validation. Compared with SurvPath, the proposed method improved the average C-index from 0.609 to 0.638, corresponding to an absolute gain of 0.029. The improvement reached significance only in BLCA under the paired \((t)\)-test, indicating cohort-dependent statistical support. Matched profiling showed lower memory consumption than the Transformer replacement, while case-level visualizations provided model-derived pathway--patch associations. These results support sample-adaptive sparse interaction modeling for WSI-omics survival prediction. External validation, repeated-seed evaluation, and pathologist-guided interpretation remain necessary before clinical application.

Research Square
Tongji University (CN), Shanghai Jiao Tong University (CN), Shanghai Qi Zhi Institute (CN), Tsinghua University (CN)
National Key Research and Development Program of China
Good health and well-being
AI in cancer detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon