Multimodal Late Fusion for Heart Failure Diagnosis Using Tabular and Chest X-ray Data: A Comparative Study of Fusion Strategies

This study develops and evaluates a multimodal artificial intelligence framework for heart failure diagnosis by integrating structured clinical data, short emergency department admission notes, and chest X-ray images. Independent models are trained for each modality, with textual notes encoded using BioClinicalBERT and chest X-rays. Predictions are combined using decision-level late fusion strategies, demonstrating improved diagnostic performance.