Fischell Fellow Spotlight: Paul Yi

Paul Yi is a Fischell Fellow. He first got involved with the Fischell Institute when he joined the University of Maryland School of Medicine as the Director of the University of Maryland Medical Intelligent Imaging (UM2ii) Center —a new radiology Artificial Intelligence Center on the Baltimore campus. Within the School of Medicine, he is an assistant professor and holds an appointment with the Department of Diagnostic Radiology and Nuclear Medicine.

Since Yi is a practicing physician-scientist, a typical day depends on whether he is on a clinical or non-clinical day. On a clinical day, Yi is in the radiology reading room interpreting imaging studies (X-rays, CT, MRI) or performing image-guided procedures such as joint injections. On a non-clinical day, Yi splits time between administrative tasks and being the principal investigator of the multidisciplinary lab that he co-directs with Vishwa Parekh, a computer scientist, and the UM2ii technical director.

Yi's areas of expertise are at the intersection of machine learning and radiology, especially in evaluating pitfalls related to machine learning in radiology, such as fairness and bias. His current focus is on assessing deep learning models' safety and trustworthiness.

This work includes identifying biases in deep learning models that may perpetuate health disparities evident in the datasets used to train them and developing solutions to mitigate them. Yi is most interested in these pitfalls of deep learning because they are often insidious but have profound implications for the safe deployment of these potentially game-changing technologies in real-world clinical practice.

In the future, Yi plans to continue building knowledge in this critical area of the intersection of machine learning with radiology. He would also like to evaluate the human-computer interaction (HCI) aspects of machine learning in radiology.

He is curious about how an artificial intelligence model's predictions impact radiologists' confidence in their diagnoses. According to Yi, understanding the human computer interaction component of machine learning will get us one step closer to making artificial intelligence a clinical reality.

Outside of work, Yi enjoys spending time with his wife and friends, serving in his church, working out, cooking, and eating good food. He also enjoys listening to podcasts and recording them for the Radiological Society of North America about artificial intelligence in radiology.

Published February 2, 2023