arXiv AI recent: Fusion is not one-size-fits-all: Cross-Modal Representation Alignment for Time-to-Event Modeling
Researchers introduced a foundation model-driven framework for cross-modal representation alignment between CT imaging and longitudinal EHR data.,The framework is designed to generalize a...
The framework uses domain-specific foundation models to encode CT and EHR modalities independently and align them in a shared latent space.,The researchers evaluated two clinically distinct TTE tasks: pulmonary embolism mortality and cardiovascular disease outcomes, on large-scale multi-instituti...