Rethinking Radiation Oncology: AI, Efficiency, and Sustainability

This study proposes a novel deep learning auto-segmentation framework guided by large language model (LLM) prompts to improve the accuracy and speed of contouring high-risk clinical tumor volumes (HR-CTV) and organs at risk (OARs) in high-dose-rate brachytherapy (HDR-BT) for cervical cancer. ChatGPT extracted data from a retrospective dataset of 32 patients with cervical cancer to generate task-specific prompts. The model achieved statistically significant improvements in segmentation metrics for all OARs (bladder, rectum, sigmoid). However, it did not improve the accuracy of HR-CTV delineation. Integrating LLM prompts with deep learning segmentation also enhances contouring consistency, minimizing inter- and intra-observer variability common in time-sensitive HDR-BT workflows.¹

This retrospective analysis by Lichter et al quantifies the annual greenhouse gas (GHG) emissions from external beam radiation therapy (EBRT) in the United States and estimates its impact on health, as well as the potential reduction in emissions via adoption of hypofractionation. This study was prompted by the recognition that health care in the United States contributes 8.5% to 10% of GHG emissions, with radiation oncology, while not an outsized contributor, does contribute significantly.² Additionally, hypofractionation is a method to provide high-quality, safe, and equitable care, with the simultaneous opportunity to decarbonize.³ Radiation oncology’s contribution of GHG comes predominantly from transportation of patients to and from visits and energy use for EBRT (notably, HVAC systems within hospitals) more so than the energy required for machines to deliver radiation itself, such as linear accelerators.⁴ In this retrospective analysis, a life cycle assessment method utilizing data on patient transportation and energy use for 556,426 patients who received EBRT for 9 cancer types in 2019-2020 was modeled using distance calculation from each patient’s zip code to the reporting hospital.

Dr Lichter’s team found that U.S. EBRT treatments resulted in 3.3 million metric tonnes of CO2e (CO2 equivalents) in 2019 and 2.7 million metric tonnes of CO2e in 2020, corresponding to 1040 and 863 disability-adjusted life-years (DALYs) lost. Considering breast cancer EBRT, which accounts for ~35% of EBRT-related emissions, Dr Lichter’s team modeled a reduction in GHG emissions from adopting hypofractionation schedules for management of breast cancer treated with the FAST-Forward protocol (5 fractions) and START-B protocol (15 fractions). With the transition from a 25-fraction regimen to 5 fractions per the FAST-Forward protocol, GHG emissions would reduce by 64%, and with the transition from a 25-fraction regimen to 15 fractions per the START-B protocol, GHG emissions would reduce by 32%. First and foremost in considering treatment paradigms is ensuring patients have access to quality, safe, and equitable care. In addition to hypofractionation providing comparable tumor control and safety outcomes to traditional schedules, Dr Lichter’s work establishes how hypofractionation can help decarbonize health care via reduction of GHG emissions.