Multi-FRuGaL: Multimodal Flexible Redundancy-aware Decomposed Gated Learning for Cancer Diagnosis and Prognosis

2026-06-04

AI Summary

This study addresses the challenge of missing or sparse multimodal cancer data commonly encountered in real-world clinical settings by proposing a decomposition-aware adaptive gating intermediate fusion framework. The method employs modality-specific encoders, a signal decomposition layer, and an input-conditioned gating mechanism to explicitly disentangle redundant and complementary information, dynamically reweighting informative modalities to suppress noise. An information-aware fusion objective function is further introduced to enhance the robustness and interpretability of learned representations. Experimental results on the HANCOCK and HECKTOR datasets demonstrate that the model achieves consistently strong performance under both complete and severely missing modalities, attaining a peak AUC of 0.975 and a concordance index of 0.7421 in survival analysis, significantly outperforming current state-of-the-art baselines.

Abstract

Modern medicine relies on heterogeneous data sources spanning radiology, pathology, text reports, and structured clinical information. However, real-world patient data are frequently incomplete, with missing or sparsely acquired modalities, limiting the effectiveness of standard multimodal fusion approaches. To this end, we propose the Multimodal Flexible Redundancy-aware decomposed GAted Learning (Multi-FRuGaL) framework, a decomposition-aware, adaptive gated intermediate-fusion framework that performs modality-level representation learning under missing data. Multi-FRuGaL integrates per-modality encoders with a signal decomposition layer, an input-conditioned gating network, and an information-aware fusion objective to separate redundant from modality-specific complementary signals, selectively upweighting informative modalities and suppressing redundant or noisy inputs, and remaining well-defined even when multiple modalities are absent. We evaluate Multi-FRuGaL on two multimodal head and neck cancer cohorts: the HANCOCK challenge dataset (N = 763) comprising five modalities and two prognostic endpoints (5-year survival and 2-year recurrence), and the HECKTOR challenge dataset (N = 588) comprising three modalities for human papillomavirus (HPV) status classification. Multi-FRuGaL consistently achieves higher mean performance than the evaluated baselines across multiple tasks, improving AUC from 0.601 to 0.8496 for survival, from 0.672 to 0.8102 for recurrence, and achieving 0.975 AUC for HPV prediction on HECKTOR. For survival analysis, it further achieves a concordance index of 0.6814 for overall survival, 0.7421 for recurrence-free survival, and 0.7143 for progression-free survival on HANCOCK, and 0.7203 for recurrence-free survival on HECKTOR. Qualitative analyses further show that Multi-FRuGaL learns discriminative and robust multimodal representations, even under severe missing-modality conditions.

Problem

• multimodal learning
• missing data
• cancer diagnosis
• data heterogeneity
• modality fusion

Innovation

• multimodal learning
• missing modality robustness
• signal decomposition
• adaptive gating
• intermediate fusion