Translating computational innovations into reality: Focus on the users!

As a professor and vice chair for research in the Department of Radiology and Imaging Sciences at Emory University School of Medicine, Elizabeth Krupinski works with medical image perception, covering human factors in radiology. Her background in experimental psychology provides her with a different perspective on computer-aided and artificial intelligence-driven tools than her colleagues with a medical background.

“I am far less interested in the actual development of artificial intelligence (AI) tools, and far more interested in how we can effectively and efficiently translate them into daily routine use in the clinic in a way that actually benefits users and patient care,” says Krupinski. “It’s great to read about an AI scheme attaining a certain level of performance — and some of them are  fine for certain clinical tasks like counting cell nuclei or labeling vertebrae in a spine image — but I would rather hear about how an AI scheme actually changed someone’s diagnostic decision so that a patient’s cancer was detected earlier, so that the right treatment started earlier, and the outcomes or patient’s survival was improved. This side of AI impact requires a very different way of looking at problems and evaluating whether or not a tool should be adopted.”

In a keynote address at SPIE Medical Imaging 2023, Krupinski, who is also the 2023 recipient of the SPIE Harrison H. Barrett Award in Medical Imaging, will provide an overview of these perspectives, drawing on the history of medical image perception research in radiology and the growing application of these principles and methods in pathology.

What are some of your responsibilities as professor and vice chair for research in the Department of Radiology and Imaging Sciences at Emory University School of Medicine?
As Vice Chair of research, I serve on the Department’s Executive Committee and interact closely with the chair, associate administrator, and other vice chairs. I serve as a key leadership resource on prioritization, direction, and strategic growth of new and existing research projects and initiatives and represent the department on appropriate school of medicine and university committees. Activities include, but are not limited to, oversight and optimization of research and research mentorship programs, strategic (re)organization of the department's research divisions, recruitment of research faculty including clinician-investigators, faculty development relevant to research, oversight of research compliance, oversight of mission-based budgeting for research operations and new programs, and facilitation of the diversification of research funding opportunities including industry collaborations and team science.

What led to your interest in medical image perception?
I was always interested in healthcare and research but got into medical image perception pretty much by chance. As an undergrad I was interested in neurobiology and behavior but gravitated more to the behavior side and how our perceptual and cognitive systems shape the way we interact with the world. My MA was on the perception of art and the aesthetic experience, but in going for my PhD I wanted to get back to healthcare and something that would have a more direct impact on the quality of people’s lives. My advisor at the time had a colleague looking for a research assistant. That happened to be Cal Nodine who was working with Hal Kundel at UPenn using eye-tracking to study the nature of errors in radiology. I got the job and that literally set my career path in motion.

In the 2019 article “Medical imaging and artificial intelligence: naturally compatible,” you noted: "Deep learning and AI are still a long way from being creative and this has been the case from the very beginning of AI implementations… As [professor of cognitive science at the University of Sussex, UK] Margarita Boden pointed out in 1998, the two major bottlenecks to AI creativity are domain expertise and evaluation of results (i.e. critical judgment of one's own original ideas)." Are these two areas still bottlenecks to AI creativity? Have there been any breakthroughs in surmounting these issues?
Sort of. AI has progressed significantly and in lots of amazing ways. It can even write papers and pass board exams, although with the former based on examples I’ve seen I’m not sure I’d give the papers much more than a C for a grade. They simply lack creativity. They present a lot of information and can follow general structures, but they’re pretty prosaic, repetitive, and not that interesting to read. AI is getting a lot closer to creativity however, so it will be interesting to see where it goes over the next few years.

What are some of the breakthroughs you’ve seen recently in your research with computer-aided/AI-driven tools?
I have not really seen any breakthroughs per se with respect to medical imaging. There have been lots of advancements and improvements, but nothing that I would call a breakthrough. Once there are AI tools being used in the clinic on an everyday basis that are having a measurable impact on the users and clinical outcomes, there will be breakthroughs that are significant.

What do you see as the future of AI in medical imaging? What would you like to see?
It will continue to develop, of course, but I would like to see more true integration of AI tools into clinical routines where they seamlessly assist users with difficult and meaningful tasks. It will not only be image-based tools that lead to change, but those that impact workflow overall: image and data acquisition, analysis, and storage; reporting; and integration with all other types of clinical data from different sources, locations, and times. Ideally all of this will work in the background synergistically with users, tailored to their individual needs, circumstances, and preferences. Another thing that needs to happen to make this all work is figuring out the return on investment so healthcare systems will buy into the multitude of systems that are going to be required to truly advance the routine clinical use of AI.

What do you want attendees to learn from your talk?
A different way of looking at and assessing AI and its impact on users, healthcare systems, and patients.