Evidential Fusion Network for Multimodal Survival Prediction under Missing Modalities

Short summary

EMMS is a novel multimodal survival prediction model designed for clinical data with missing modalities, using Dempster-Shafer theory and Gaussian Random Fuzzy Numbers to quantify uncertainty. It achieves state-of-the-art performance on four cancer datasets without requiring generative imputation. Applicable for data scientists handling incomplete multimodal healthcare observations.

• •EMMS handles missing modalities through Dempster-Shafer uncertainty fusion, treating gaps as vacuous evidence
• •SOTA performance on 4 cancer datasets with calibrated uncertainty estimates and no added computational overhead
• •Avoids generative imputation for missing data, reducing complexity in clinical ML workflows

Generated with AI, which can make mistakes.