Trusted CI Webinar: Securing Medical Imaging AI Models Against Adversarial Attacks

While AI is increasingly present in clinical practice especially for medical imaging, it is imminent to ensure cybersecurity of imaging diagnostic AI models. Newly advanced adversarial attacks pose a threat to the safety of medical AI models, but little is known about the characteristics of this threat. Medical adversarial attacks may lead to serious consequences including patient harm, liability of healthcare providers, and other ethical issues or crimes. It is imperative to study this cybersecurity issue to mitigate potential negative consequences and to ensure safety of health care. In this talk, the speaker will discuss cyber vulnerabilities of deep learning-based medical imaging diagnosis models under adversarial attacks, show real-world experiments on how adversarial attacks can fool AI models to decrease diagnosis performance and to confuse experienced radiologists, and present several methods of defending adversarial attacks to secure AI models in medical imaging applications. Speaker Bio: Shandong Wu, PhD, is a Professor in Radiology, Biomedical Informatics, Bioengineering, and Intelligent Systems at the University of Pittsburgh. Dr. Wu leads the Intelligent Computing for Clinical Imaging (ICCI) lab, and he is the founding director of the Pittsburgh Center for AI Innovation in Medical Imaging. Dr. Wu’s work focuses on developing trustworthy medical imaging AI for clinical/translational applications. Dr. Wu's lab received multiple research awards such as the RSNA Trainee Research Award twice in 2017 and 2019, the 2021 AANS Natus Resident/Fellow Award for Traumatic Brain Injury, the 2025 SPIE Imaging Informatics Best Paper Award, etc. Dr. Wu’s research is supported by NIH, NSF, multiple research foundations, Amazon AWS, Nvidia, and many institutional funding sources. Dr. Wu has published > 190 journal papers and conference papers/abstracts in both the computing and clinical fields. His research has been featured in hundreds of scientific news reports and media outlets in the world. This material is based upon activities supported by the National Science Foundation under Interagency Agreement #A2407-049-089-064206.0. Any opinions, findings, and conclusions or recommendations expressed are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.