AI, Team Science, and Emory Radiology: Transformative Innovations

When people think about radiologists, they typically imagine a physician in a dark room, peering at a screen image, searching for signs of disease. At Emory, that notion of an isolated figure in a reading room is changing. The era of the lone-wolf radiologist is over. Artificial intelligence (AI) has begun transforming medical imaging, as machine learning offers new opportunities for processing quantitative data and supporting human health. AI’s ability to recognize patterns makes it an ideal match for imaging sciences. But AI isn’t the only reason radiology is changing. The real story is the team science making the clinical advances possible—and connecting quantitative imaging to justice, diversity, and equity commitments across Emory healthcare and beyond.

Who are these scientists? Led by faculty in Emory’s Department of Radiology and Imaging Sciences, this interdisciplinary cohort is using machine learning to rethink how we diagnose and treat disease—and how to do so equitably. Faculty engaged in this cooperative effort bring expertise in interventional radiology, computer science, natural language processing, population health, emergency medicine, ethics, and healthcare operations, among other fields. Together with industry collaborators and internal and external academic partners, these clinician scientists are catapulting Emory into the forefront of machine learning in medicine. They’re doing so not just for imaging sciences, but also for healthcare as a whole.

“People are surprised by how broad radiology’s focus can be,” says Elizabeth Krupinski, professor and vice chair of research in radiology. “It’s not just about imaging.” Today’s radiologists tend to be technologically savvy, comfortable with rapid change, and sensitive to the demands of clinical settings. They understand that embracing quantitative imaging doesn’t mean losing sight of the individual patient. Instead, it means creating more personalized approaches to care, grounded in the latest innovations and attentive to questions of bias and medical fairness.

Judy Gichoya, assistant professor of radiology, recently conducted a collaborative study showing how AI can recognize patients’ racial identities in medical images. A computer’s ability to predict self-reported race raises important questions. How do we navigate the phenomenon of bias in artificial intelligence models, especially as machine learning moves into the mainstream of healthcare and clinical operations?

To address such questions, the radiology team has partnered with Emory’s Center for Ethics. Should AI be used at all within medical imaging? If so, how can healthcare leaders ensure equitable access? What disparities or biases might machine learning help address? What disparities or biases might it exacerbate? Medical ethicist John Banja has launched a grant-funded podcast series, “AI, Radiology and Ethics,” that explores these challenges in conversation with Emory experts.

The ethics podcast is just one of the collaborations underway in the radiology team science arena. A second involves Emory’s HITI Lab (Healthcare Innovations and Translational Informatics), which aims to apply machine learning to big problems in healthcare and diagnostics. To date, lab members have brought in more than $1.6 million in grants, with one recent project examining the performance of intensive care unit severity scoring systems across different ethnicities in the United States. Other projects include leading large research collaboratives on medical AI, hiring joint computer sciences faculty, and contributing to the AI ecosystem at Emory through academic industry partnerships while also training a diverse workforce of up-and-coming imaging scientists.

Judy Gichoya, Assistant Professor of Radiology

Fall 2021 lab retreat with students from Emory, Georgia Tech, and Kennesaw State university

The team science approach is also alive and well in Emory’s Translational Laboratory for Cardiothoracic Imaging and Artificial Intelligence. Founded in 2020, this collaboration aims to perform advanced imaging research using the latest technologies, with special emphasis on clinical validation, integration, and implementation. Led by Carlo De Cecco, associate professor of radiology and biomedical informatics, lab members are exploring how machine learning aids radiologists’ ability to predict a cardiac event. They are also examining AI as a tool for pulmonary disease severity classification and prognostication, combining demographic data, lab values, medication data, medical history, and socioeconomic factors.

Both in the lab and outside it, Emory radiology faculty remain committed to patients and the patient experience. Gichoya and Krupinski have each received National Science Foundation grants in partnership with the University of Indiana to examine clinical and equity issues surrounding the implementation of AI in healthcare settings. “A lot of groups don’t do this part,” says Krupinski. “We’re taking our work into the clinic and looking at AI from the perspective of the users.”

After all, if a healthcare provider can’t make timely and efficient use of the new data available, or if the patient can’t access a needed procedure or timely care, then more change is needed.

“As AI continues to be integrated in our everyday lives, the issue of impact on society is increasingly important,” says Gichoya. “For us working in AI in medicine, we have a lot of work to do, to make sure our patients remain safe, and that we all reap the AI benefits. Our team here at Emory is increasingly relied on for our growing expertise on fairness and bias, as well as the curation of diverse and representative datasets.”

Led by principal investigator and radiologist Carlo N. De Cecco MD, PhD, the Translational Laboratory for Cardiothoracic Imaging and Artificial Intelligence develops, evaluates, and implements novel cardiothoracic imaging techniques, analysis methods such as radiomics, and artificial intelligence applications.

“People are attracted to radiology who want to embrace that progress. It’s an exciting time to be in medicine.”

Caroyln Meltzer

Carolyn Meltzer, William P. Timmie Professor and Chair of Radiology, adds that radiology tends to operate on the technological cutting edge of medicine, with imaging scientists poised to have a substantial impact across specialties and disease systems. “People are attracted to radiology who want to embrace that progress,” she says. “It’s an exciting time to be in medicine.”

And while Nabile Safdar, vice chair for imaging informatics at Emory, acknowledges that people sometimes ask him if radiologists will be out of a job in five or ten years because of advances in artificial intelligence, the reality, he says, is the opposite. “AI is good about giving specific answers to specific questions,” he says. “But it’s not good at asking the questions that need to be asked. Radiologists understand what questions need to be asked.”

Pursuing knowledge at the intersection of clinical workflow, front-end user experience, and technical processes, the team scientists at Emory radiology are improving disease detection and treatment—all while seeking greater equity and access for every patient.

Carolyn Meltzer, William P. Timmie Professor and Chair of Radiology

Adam Prater, Department of Radiology and Imaging SciencesAssistant Professor Nabile Safdar, Vice Chair for Imaging Informatics at Emory