Radiation and Cancer Physics

EDU 50 - Interpretability and Uncertainty Quantification: A Way toward Trustworthy AI in Radiation Oncology

Moderator(s)

• 

  Lei Ren, PhD

  University of Maryland Radiation Oncology
  Baltimore, MD, United States

AI has gained significant advances in recent years, and started to be implemented for a broad range of applications in radiation therapy, including imaging, segmentation, treatment planning and delivery, etc. Various studies, mostly carried out through retrospective analysis, demonstrated the potential of AI to improve both the efficiency and accuracy of radiation therapy. Recent breakthroughs in Artificial General Intelligence (AGI) powered by Large Language Models (LLM) further boosted the development and adoption of AI in clinical practices.
Despite the tremendous progress in AI development, many practical challenges exist when AI is applied prospectively in the clinical practice. AI is often treated as a black box without clear understanding of its decision process. This lack of interpretability leads to concerns about the robustness and generalizability of the models, which hinders clinicians confidence in the models when using them in daily practice.
This session will focus on the aspects of trustworthy AI in radiation oncology. Specifically, it will introduce and discuss the progress and future directions of two areas regarding developing trustworthy AI: (1) Interpretable AI. Significant progresses have been made in developing interpretable models to increase AI transparency while maintaining the performance. (2) Uncertainty quantification: Novel methods have been developed and implemented to estimate the uncertainties in AI predictions, which are critical to designing safeguards to minimize the risk of AI decision in clinical practice. Speakers will review the latest developments in these two directions, and have a panel discussion about the future roadmap for achieving trustworthy AI in medicine.

Presentations:

• 4:00 PM – 4:02 PM ET

  Introductions
  

  Speaker: Lei Ren, PhD – University of Maryland Radiation Oncology
• 4:02 PM – 4:17 PM ET

  Interpretable AI: Advancements, Challenges and Paths Ahead
  

  Speaker: Lei Ren, PhD – University of Maryland Radiation Oncology
• 4:17 PM – 4:32 PM ET

  AI Uncertainty Estimation: Approaches and Implementations
  

  Speaker: Xiaofeng Yang, PhD – Emory University
• 4:32 PM – 4:47 PM ET

  Finding Clarity in the "Black-Box" of AI and Integration into Clinical Care
  

  Speaker: Manisha Palta, MD – Duke Cancer Institute
• 4:47 PM – 5:00 PM ET

  Questions and Answers
  

  Speaker: Lei Ren, PhD – University of Maryland Radiation Oncology