Continued from cover
At yesterday’s plenary session, Judy Wawira Gichoya, MD, MS, FSIIM, explored why AI adoption lags behind its theoretical promise, drawing on high-profile cases in radiology, pathology, and beyond. In a talk entitled, "Artificial intelligence in real world clinical settings," Gichoya discussed high-profile failings from clinical prediction models for sepsis, presenting common underpinnings for these barriers and more. Gichoya is co-director of Emory University’s Healthcare AI Innovation and Translational Informatics (HITI) Lab.
At the heart of the excitement in this space is the innovation that enables generalizable applications. Gichoya provided an informative overview of the background and fundamentals driving AI systems before demonstrating the awe-inspiring capabilities of state-of-the-art foundation models and multi-modal systems. These publicly available, easy-to-navigate systems allow for fascinating explorations of image-to-text, image-to-speech, and text-to-text transformations, among many more.
In shifting to a more practical view of how these technologies may impact medicine, she walked us through the many caveats exist in real-world implementations. Through learning from others' experiences, she presented a crucial set of considerations for laboratorians as we wade into the AI waters. Not the least of which includes evaluating how AI impacts model performance, algorithmic fairness, implementation barriers, patient advocacy, and much more.
This insightful and informative session provided laboratorians and clinicians alike with practical strategies to rigorously evaluate AI models in clinical settings. Gichoya looked toward the future for how we might streamline these elusive implementation efforts more broadly.
