Three discipline collaborative radiation therapy (3DCRT) special debate: AI structure segmentation is better than clinician contouring for both OARs and targets

Abstract

Radiation Oncology is a highly multidisciplinary medical specialty, drawing significantly from three scientific disciplines—medicine, physics, and biology. As a result, discussion of controversies or changes in practice within radiation oncology involves input from all three disciplines, and sometimes more! For this reason, significant effort has been expended recently to foster collaborative multidisciplinary research in radiation oncology, with substantial demonstrated benefit. In light of these results, we endeavor here to adopt this “team-science” approach to the traditional debates featured in this journal. This article is part of the series of special JACMP debates entitled “Three Discipline Collaborative Radiation Therapy (3DCRT)” in which each debate team typically includes a radiation oncologist, a medical physicist, and a radiobiologist. In this case, we have included a medical dosimetrist. We hope that this format will not only be engaging for the readership but will also foster further collaboration in the science and clinical practice of radiation oncology and developments thereof.

Artificial intelligence (AI) is ubiquitous. The applications are limitless and the effects are permeative. The use of AI for contouring of organs at risk (OARs) has been in the works now for many years, however as algorithms have improved and adaptive replanning is becoming increasingly prevalent in the clinic, physicians and radiation oncology teams are increasingly reliant on software for auto contouring, including in certain scenarios for contouring targets. In this debate, we consider the risks and benefits of this progression towards increased contouring by AI. At what point does the machine definitively outperform the clinician? Are we there yet? For this debate we will argue exactly this point through the proposition, “AI structure segmentation is better than clinician contouring for both OARs and targets.” Arguing for the proposition will be John Kang, Stine Korreman, Brian Napolitano, and Sharif Elguindi. John Kang, MD, PhD, is an assistant professor in radiation oncology in the University of Washington Department of Radiation Oncology and the Fred Hutch Cancer Center. He is dual board certified in radiation oncology and clinical informatics and serves as clinical informatics lead. His clinical focus is on thoracic malignancies and his research focus is on natural language processing and informatics applications. Stine S Korreman, PhD, is Professor of Medical Physics at Aarhus University, Denmark. She leads a research group on AI for medical image analysis in radiotherapy with a focus on segmentation and dose prediction, and translation from research to clinical practice. She is chair of the ESTRO Focus Group AI in Radiotherapy and Director of the ESTRO course on AI in Radiotherapy. Brian Napolitano, MHL, CMD is Director of Medical Dosimetry at Massachusetts General Hospital in Boston, where he oversees treatment planning operations for photon and proton modalities at their main campus and satellite facilities. Brian is a former president of the American Association of Medical Dosimetrists (AAMD) and was the 2024 recipient of the AAMD Outstanding Achievement Award.  He received his Bachelor of Science degree in Biological Sciences from Binghamton University and his Master of Healthcare Leadership degree from Brown University.  Lastly, Sharif Elguindi, MS, DABR, is a Medical Physicist at Memorial Sloan Kettering Cancer Center where he serves as the AI clinical implementation lead. Together with his team, he has helped design, develop, and maintain an AI-assisted contouring system that improves workflow efficiencies for over 10 000 treatments annually. His professional interests focus on designing and implementing software systems that support AI-assisted target workflows for physicians.

Arguing against the propostion will be Andrew Hope, Michelle Mundis, and Jan-Jakob Sonke. Andrew Hope, MD, FRCPC, is a Clinician Investigator in the Radiation Medicine Program, Princess Margaret Cancer Centre, Associate Professor in the Department of Radiation Oncology at University of Toronto and the Addie MacNaughton Chair in Thoracic Radiation Oncology. His research focuses on developing, deploying, and evaluating novel AI applications and other advanced technologies in the clinic. Michelle Mundis, MS, CMD, is a senior dosimetrist at Maryland Proton Treatment Center (MPTC). She has over 10 years of experience in radiation oncology, including roles as field service engineer, Varian Medical Systems, Medical Physics Assistant, and Clinical Coordinator for the University of Maryland Dosimetry Program. She currently serves as Secretary on the Board of Directors of the American Association of Medical Dosimetrists. Jan-Jakob Sonke, PhD, leads a research group on adaptive radiotherapy at the Netherlands Cancer Institute (NKI). He is also the theme lead for image guided therapy at the NKI and full professor at the University of Amsterdam. He is one of the scientific directors of two labs focusing on the development of innovative AI algorithms for oncology and radiation therapy.

The first seven authors contributed equally to this work. All authors were responsible for preparation of arguments and in writing and reviewing the manuscript.

Sharif Elguindi has two patents pending, one on an AI foundation model for 3D medical image segmentation and one for AI inference and quality assurance software.