Embracing Aleatoric Uncertainty in Medical Multimodal Learning with Missing Modalities

TL;DR

This paper introduces Aleatoric Uncertainty Modeling (AUM) for medical multimodal learning, explicitly quantifying uncertainty to improve handling of missing modalities and enhance predictive performance in clinical tasks.

Contribution

The paper proposes a novel AUM framework that models unimodal uncertainty and uses a dynamic, uncertainty-aware aggregation mechanism for better multimodal data integration.

Findings

Achieved 2.26% AUC-ROC improvement on MIMIC-IV mortality prediction.

Gained 2.17% AUC-ROC improvement on eICU dataset.

Outperformed existing state-of-the-art methods in multimodal medical prediction.

Abstract

Medical multimodal learning faces significant challenges with missing modalities prevalent in clinical practice. Existing approaches assume equal contribution of modality and random missing patterns, neglecting inherent uncertainty in medical data acquisition. In this regard, we propose the Aleatoric Uncertainty Modeling (AUM) that explicitly quantifies unimodal aleatoric uncertainty to address missing modalities. Specifically, AUM models each unimodal representation as a multivariate Gaussian distribution to capture aleatoric uncertainty and enable principled modality reliability quantification. To adaptively aggregate captured information, we develop a dynamic message-passing mechanism within a bipartite patient-modality graph using uncertainty-aware aggregation mechanism. Through this process, missing modalities are naturally accommodated, while more reliable information from available modalities is dynamically emphasized to guide representation generation. Our AUM framework achieves an improvement of 2.26% AUC-ROC on MIMIC-IV mortality prediction and 2.17% gain on eICU, outperforming existing state-of-the-art approaches.