ECE 290 Seminar: From Code to Clinic: How Regulatory Science and Virtual Trials Ensure Trustworthy AI in Medical Imaging

Presenter: Dr. Brandon Nelson, Staff Fellow, Division of Imaging, Diagnostics, and Software Reliability (DIDSR), U.S. Food and Drug Administration’s Center for Devices and Radiological Health (CDRH)

Description: Artificial intelligence is rapidly transforming diagnostic and interventional radiology, presenting immense opportunities for improving patient care alongside significant regulatory challenges. As AI/ML-enabled devices proliferate, how do we ensure they are safe, effective, and perform robustly across the diverse patient populations and imaging hardware seen in clinical practice?

This talk will introduce the FDA’s role in device evaluation and address the urgent need for new scientific methods to keep pace with the rapid evolution of AI-based technologies. The presentation will focus on the critical role of regulatory science—the discipline of creating novel tools and methodologies to assess the safety, efficacy, and performance of these complex, data-driven systems.

I will detail my research at the FDA, which centers on developing and applying innovative in silico evaluation tools, specifically virtual clinical trials that leverage synthetic patient data and simulated pathologies. These tools are critical for addressing real-world data gaps and allow for the rigorous, systematic "stress-testing" of AI algorithms. I will share specific examples from my research, including the creation of virtual cohorts to assess AI robustness for tasks like intracranial hemorrhage detection and CT image reconstruction, with a particular focus on ensuring generalizability to underrepresented pediatric populations.

Finally, I will discuss how these regulatory science frameworks can create powerful collaborative opportunities, bridging the gap between cutting-edge academic innovation in areas like photon-counting and C-arm CT and the evidence needed to bring trustworthy AI safely to the clinic.

Bio: Dr. Brandon Nelson is a Staff Fellow within the Division of Imaging, Diagnostics, and Software Reliability (DIDSR) at the U.S. Food and Drug Administration’s Center for Devices and Radiological Health (CDRH). In this unique role, he combines regulatory duties with scientific research, serving as a subject matter expert and reviewer for medical device submissions that incorporate AI/ML and advanced image reconstruction, while also leading research to develop novel tools for their evaluation.

His primary research focuses on leveraging regulatory science—including in silico methods like virtual clinical trials and synthetic data generation—to assess the robustness, generalizability, and safety of medical imaging AI, with a special emphasis on addressing performance gaps in pediatric populations.

He earned his Ph.D. in Biomedical Engineering from the Mayo Clinic Graduate School of Biomedical Sciences, where his dissertation focused on multi-contrast, grating-based interferometry for micro-CT. Dr. Nelson is a recipient of the FDA's Critical Path Award for his work on pediatric AI evaluations and has published extensively in leading medical physics journals and conferences.

Hosted by: Professor Soumya Bose, ECE Department