中央研究院 Logo

The rapid expansion of medical data has created new opportunities to redefine disease characterization beyond traditional clinical measurements. This lecture introduces an AI-enabled deep phenotyping framework that leverages electrocardiograms, chest radiographs, and electronic health records to derive novel digital biomarkers. By integrating multimodal learning and foundation models, artificial intelligence can extract latent physiological information, predict a wide range of clinical phenotypes, and reveal biologically meaningful, non-random prediction errors that extend current disease taxonomies. Beyond model development, this talk emphasizes the importance of evaluating AI as a clinical intervention. We demonstrate how AI-derived biomarkers can be embedded into real-world workflows through pre-specified clinical actions and assessed using pragmatic randomized clinical trials to establish causal effects on patient outcomes. Finally, real-world deployments in community screening and home-based care are presented to illustrate how AI can function as a “digital microscope,” connecting hospital and out-of-hospital data to enable scalable, evidence-based precision medicine.