Development of a Multimodal Multi-Task Artificial Intelligence System for Early Prognostic Assessment, Infection Identification, and Pathogen Guidance in Critically Ill ICU Patients

50000

Patients in the Intensive Care Unit (ICU) are critically ill with rapidly changing conditions. Accurate prognosis assessment and timely identification of infectious etiology represent major clinical challenges. Existing scoring systems rely on static data and cannot capture dynamic disease evolution, while traditional microbiological cultures are time-consuming and have low positive rates. The breakthrough of this project highly depends on the deep cross-disciplinary integration and complementary strengths of AI-driven big data analytics and clinical critical care medicine. The XJTLU engineering team focuses on model building, leveraging expertise in computer vision and data science to construct a dual-channel multi-task deep learning architecture; the Ruijin medical team focuses on utilizing high-quality single-center retrospective time-series clinical data and imaging features to reveal the clinical value of the AI model in prognostic stratification and infection etiology guidance. Through close collaboration, the two teams will innovatively integrate the first chest X-ray (CR) of ICU patients with time-series data collected within 24 hours of admission, achieving a "three-in-one" clinical decision support tool: predicting 28-day all-cause mortality, identifying infection, and providing initial guidance on pathogen category. The project outcomes will provide ICU clinicians with a rapid, non-invasive decision-support tool to optimize early empirical anti-infective therapy, with highly promising clinical application prospects.