FDB (First Databank, Inc.)’s Post

Generative AI (GenAI) has performed surprisingly well on numerous healthcare-related tests, even passing U.S. medical licensing exams. A new study, however, shows that the technology may need more training in forming diagnoses based on clinical images. The National Institutes of Health researchers asked a GenAI model to review 207 clinical images and associated questions and then provide a written rationale for its chosen diagnosis. While the GenAI scored better than a physician control group in a “closed book” setting without any outside resources, it often made mistakes describing images and explaining its reasoning for the chosen diagnosis. Researchers believe that the study shed new light on GenAI’s abilities and limitations in analyzing images and demonstrated how crucial it is for healthcare organizations to test new AI models before introducing them into practice. https://lnkd.in/gpwRdh3u | #AI #AIinHealthcare #ClinicalDecisionSupport #CDS #NIH