Artificial intelligence can overcome challenges in brachytherapy treatment planning

Abstract

In the field of radiotherapy (RT) in recent years, there has never been any diminution in enthusiasm of adopting artificial intelligence (AI), in the form of deep learning and/or machine learning, mainly for external beam treatment applications. A PubMed search with the keywords “artificial intelligence” and “radiotherapy” would return thousands in publications, whereas if the keywords are changed to “artificial intelligence” and “brachytherapy,” results would end up in lower hundreds. The interest in applying AI toward the brachytherapy planning process, including applicator digitization, contouring, plan optimization, and so forth has never fully blossomed. The reasons might be mainly twofold: (1) AI requires large amount of uniform training data, while “brachytherapy” might be the antonym of “uniform” and “large data,” considering the treatment variations and case amount at each institution; (2) each treatment planning step is straightforward but poses unique challenges, considering the use of multi-modality images, image artifacts from applicators, case-specific dose distribution, and so forth. Taking a walk down the memory lane, brachytherapy treatment planning has been advancing slowly in general compared to external beam RT, going from standard pear-shape planning not too long ago to MRI-guided high-risk clinical planning volume gold-standard in the recent years. When considering devoting limited resources in further advancing brachytherapy, should we take such a big leap to AI or should we take a conservative route exploring more traditional computation methods? In other words, do we have the faith that AI can ultimately overcome those challenges in brachytherapy and provide better solutions? In this debate, Dr. Xun Jia argues for the proposition that “AI can overcome challenges in brachytherapy treatment planning,” while Dr. J. Adam M. Cunha argues against it. Dr. Xun Jia is Professor and Associate Vice Chair of Medical Physics Research at the Department of Radiation Oncology, University of Texas Southwestern Medical Center (UTSW). He received his master's degree in applied mathematics in 2007 and Ph.D. degree in physics in 2009, both from the University of California Los Angeles. After his postdoctoral training in medical physics from the Department of Radiation Physics and Applied Sciences, University of California San Diego, he became a faculty in the same department in 2011. He moved to UTSW in 2013. Over the years, Dr. Jia has conducted productive research on cone beam CT reconstruction, GPU-based Monte Carlo radiation transport simulation, deep learning for image processing and RT treatment planning, and development of a preclinical small animal radiation research platform. He has published ∼150 peer-reviewed manuscripts. His research has been funded by NIH, the State of Texas, industrial, and charitable funding agencies. Dr. Jia currently serves as an Executive Editorial board member of Physics in Medicine and Biology. He is the recipient of John Laughlin Young Scientist Award of American Association of Physicists in Medicine in 2017. Dr. J. Adam M. Cunha is a resident of the San Francisco Bay Area in Northern California. He is an Associate Professor in the Department of Radiation Oncology at the University of California, San Francisco. Dr. Cunha first started working with machine learning in 2002 where it was a crucial part of his Ph.D. thesis searching for rare sub-atomic particles at the SLAC National Laboratory. Since transitioning into medical physics Dr. Cunha has dedicated his career to improving brachytherapy practice. He has focused his research on optimization and hardware development including robotic devices, electromagnetic tracking, 3D printing, and treatment planning algorithms. Brachytherapy treatment planning generally encompasses steps of imaging, structure delineation, dose calculation, plan optimization, and so forth. The goal here is to accurately accomplish these steps in a timely fashion.