Yonsei University develops multimodal medical AI for incomplete clinical data

Researchers at Yonsei University College of Medicine have developed a new multimodal medical AI framework called ‘ETF-UML’. This system is designed to function reliably in real-world clinical settings where patient data, such as laboratory results or clinical records, may be incomplete or missing.

Unlike existing models that often require complete datasets or rely on potentially inaccurate methods to fill in missing information, ETF-UML allows different types of data—including tabular, time-series, and text—to share a common decision-making standard from the start. In evaluations using international databases like MIMIC-IV and eICU, the framework demonstrated superior performance in predicting patient mortality compared to existing models, particularly when data availability was low.

Additionally, the framework is computationally efficient, requiring significantly less graphic memory and processing power than previous multimodal AI models.