April 11, 2024
Paul Yi, MD | University of Maryland School of Medicine
Although AI and deep learning have demonstrated tremendous potential to perform expert-level radiology diagnostic tasks, these same tools can also demonstrate bias, potentially perpetuating pre-existing health inequities. In this talk, we will review practical considerations for addressing fairness and bias issues in AI for radiology. Topics to be discussed include how to optimally perform dataset curation, the need for precision in the definitions of bias used to evaluate AI tools in radiology, and the importance of performance monitoring after clinical deployment. Although the use cases discussed will be specific to radiology, the principles and concepts will apply to AI in any clinical specialty.
