Washington University in St. Louis is set to become the second institution with a doctoral program dedicated to imaging science, with its inaugural class beginning in the fall. Mark Anastasio, PhD, the director of the new program, spoke with Health Imaging about the state of imaging education and incorporating artificial intelligence (AI) into its curriculum.
Health Imaging: Where did the idea for this PhD program start? How did it get off the ground?
Mark Anastasio, PhD: I’ve been interested in pushing imaging science as an independent academic discipline for a while, but I didn’t come up with this idea alone. There are giants in the field of imaging science who have advocated this for decades. At Washington University, it came about primarily from the dean of the engineering school, Aaron Bobick, who has a background in computer vision—so he’s familiar with imaging technologies. He and I decided this would be a great opportunity for Washington University to take leadership on the educational aspects of imaging science.
The PhD program at the Rochester Institute of Technology is the only other imaging science degree offered in the country. What makes your program different?
Our program has a distinct focus on the computational and mathematical principles of modern imaging science. We’re focused on the quantitative and computational aspects, where the Rochester program I think is more broad-based. There’s a strong radiology department and medical school here and we have a rich history of imaging science. PET technology was pioneered here along with other medical imaging technologies. This environment will probably be richer in terms of medical imaging applications.
What is the general state of medical imaging right now? Are there enough young people being educated for the increasing demand in imaging specialists?
There continues to be a healthy demand for PhD level imaging scientists. In the professional industry there’s this renaissance in artificial intelligence technology and deep learning. These are rapidly infiltrating the medical imaging world. Scientists who are engineers trained in modern methodologies, such as machine learning, but also understand imaging physics and image formation in principle will be very well positioned to make important contributions.
How will you incorporate AI into your program?
As part of our core curriculum we have one course dedicated to data-driven imaging, which will provide an introduction to how machine learning technologies and imaging science can be naturally coupled.
What advantages will students trained on machine learning and AI in programs such as this have over those who aren’t?
Learning how these AI technologies can be used within imaging science problems is very important and empowering. The students in our program will receive training in classical topics such as imaging physics, physical properties of images and how to form images, but they’ll also learn data-driven methodologies that exploit these new powerful AI methodologies. It’s a broad skill set I think students who are traditionally trained wouldn’t necessarily have.
Do you notice any generational gravitation toward AI and machine learning?
I’ve definitely noticed that. It’s kind of a hot topic. Everyone wants to study machine learning, but especially the younger people.
There’s widespread fear that AI will take over many positions in radiology and imaging. Do you believe we’ll need more or fewer imaging specialists in the future?
My personal opinion is that AI technologies will not put all radiologists out of jobs. I think it will change the way in which radiologists operate and those who embrace these technologies will be more efficient and more accurate in diagnoses.
Do students in your program and researchers in your field have a role to play in creating a smooth transition to AI and machine learning?
Certainly, we’d like to contribute to a smooth adaptation of new technologies. When we develop new technologies, our hope is they’ll be used clinically and improve outcomes for patients. Although, we’re also interested in being mindful about what makes sense and what doesn’t.
The way that AI technologies should be used within a medical imaging context might be different, for example than if you’re using AI for pure traditional computer vision problems. The consequences for making a correct or incorrect decision is very different when you’re dealing with medical imaging than a non-medical application.
When we develop or adopt new technologies for medical imaging purposes we have to be a little careful and maybe even more careful than people who are outside the medical imaging sector.
Are there any AI-related technologies you are particularly excited about?
The popular topic is computer-aided diagnoses—using deep learning to help interpret images in various ways. It’s a promising area. I’m also interested in the potential use of AI tools in other areas of imaging science such as image reconstruction and image formation. That’s basically capturing information gathered in a database of images which could potentially be useful for improving image reconstruction.
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.