Purpose: We present the first study to investigate Large Language Models (LLMs) in answering radiation oncology physics questions. Because popular exams like AP Physics, LSAT, and GRE have large test-taker populations and ample test preparation resources in circulation, they may not allow for accurately assessing the true potential of LLMs. This paper proposes evaluating LLMs on a highly-specialized topic, radiation oncology physics, which may be more pertinent to scientific and medical communities in addition to being a valuable benchmark of LLMs.Methods: We developed an exam consisting of 100 radiation oncology physics questions based on our expertise. Four LLMs, ChatGPT (GPT-3.5), ChatGPT (GPT-4), Bard (LaMDA), and BLOOMZ, were evaluated against medical physicists and non-experts. The performance of ChatGPT (GPT-4) was further explored by being asked to explain first, then answer. The deductive reasoning capability of ChatGPT (GPT-4) was evaluated using a novel approach (substituting the correct answer with “None of the above choices is the correct answer.”). A majority vote analysis was used to approximate how well each group could score when working together.Results: ChatGPT GPT-4 outperformed all other LLMs and medical physicists, on average, with improved accuracy when prompted to explain before answering. ChatGPT (GPT-3.5 and GPT-4) showed a high level of consistency in its answer choices across a number of trials, whether correct or incorrect, a characteristic that was not observed in the human test groups or Bard (LaMDA). In evaluating deductive reasoning ability, ChatGPT (GPT-4) demonstrated surprising accuracy, suggesting the potential presence of an emergent ability. Finally, although ChatGPT (GPT-4) performed well overall, its intrinsic properties did not allow for further improvement when scoring based on a majority vote across trials. In contrast, a team of medical physicists were able to greatly outperform ChatGPT (GPT-4) using a majority vote.Conclusion: This study suggests a great potential for LLMs to work alongside radiation oncology experts as highly knowledgeable assistants.

Intensity modulated arc therapy (IMAT) has been widely adopted for Stereotactic Body Radiotherapy (SBRT) for lung cancer. While treatment dose is optimized and calculated on a static Computed Tomography (CT) image, the effect of the interplay between the target and linac multi-leaf collimator (MLC) motion is not well described and may result in deviations between delivered and planned dose. In this study, we investigated the dosimetric consequences of the inter-play effect on target and organs at risk (OAR) by simulating dynamic dose delivery using dynamic CT datasets. Fifteen stage I non-small cell lung cancer (NSCLC) patients with greater than 10 mm tumor motion treated with SBRT in 4 fractions to a dose of 50 Gy were retrospectively analyzed for this study. Each IMAT plan was initially optimized using two arcs. Simulated dynamic delivery was performed by associating the MLC leaf position, gantry angle and delivered beam monitor units (MUs) for each control point with different respiratory phases of the 4D-CT using machine delivery log files containing time stamps of the control points. Dose maps associated with each phase of the 4D-CT dose were calculated in the treatment planning system and accumulated using deformable image registration onto the exhale phase of the 4D-CT. The original IMAT plans were recalculated on the exhale phase of the CT for comparison with the dynamic simulation. The dose coverage of the PTV showed negligible variation between the static and dynamic simulation. There was less than 1.5% difference in PTV V95% and V90%. The average inter-fraction and cumulative dosimetric effects among all the patients were less than 0.5% for PTV V95% and V90% coverage and 0.8 Gy for the OARs. However, in patients where target is close to the organs, large variations were observed on great vessels and bronchus for as much as 4.9 Gy and 7.8 Gy. Limited variation in target dose coverage and OAR constraints were seen for each SBRT fraction as well as over all four fractions. Large dose variations were observed on critical organs in patients where these organs were closer to the target.

Multiple energy extraction (MEE) is a technology that was recently introduced by Hitachi for its spot-scanning proton treatment system, which allows multiple energies to be delivered in a single synchrotron spill. The purpose of this paper is to investigate how much beam delivery time (BDT) can be reduced with MEE compared with single energy extraction (SEE), in which one energy is delivered per spill. A recently developed model based on BDT measurements of our synchrotron's delivery performance was used to compute BDT. The total BDT for 2694 beam deliveries in a cohort of 79 patients treated at our institution was computed in both SEE and 9 MEE configurations to determine BDT reduction. The cohort BDT reduction was also calculated for hypothetical accelerators with increased deliverable charge and compared with the results of our current delivery system. A vendor-provided MEE configuration with 4 energy layers per spill reduced the total BDT on average by 35% (41 seconds) compared with SEE, with up to 65% BDT reduction for individual fields. Adding an MEE layer reduced the total BDT by <1% of SEE BDT. However, improving charge recapture efficiency increased BDT savings by up to 42% of SEE BDT. The MEE delivery technique reduced the total BDT by 35%. Increasing the charge per spill and charge recapture efficiency is necessary to further reduce BDT and thereby take full advantage of our MEE system's potential to improve treatment delivery efficiency and operational throughput.

Purpose: To quantitatively investigate the effect of range shifter materials on single‐spot characteristics of a proton pencil beam.Methods: An analytic approximation for multiple Coulomb scattering (“differential Moliere” formula) was adopted to calculate spot sizes of proton spot scanning beams impinging on a range shifter. The calculations cover a range of delivery parameters: six range shifter materials (acrylonitrile butadiene styrene, Lexan, Lucite, polyethylene, polystyrene, and wax) and water as reference material, proton beam energies ranging from 75 to 200 MeV, range shifter thicknesses of 4.5 and 7.0 g/cm², and range shifter positions from 5 to 50 cm. The analytic method was validated by comparing calculation results with the measurements reported in the literature.Results: Relative to a water‐equivalent reference, the spot size distal to a wax or polyethylene range shifter is 15% smaller, while the spot size distal to a range shifter made of Lexan or Lucite is about 6% smaller. The relative spot size variations are nearly independent of beam energy and range shifter thickness and decrease with smaller air gaps.Conclusions: Among the six material investigated, wax and polyethylene are desirable range shifter materials when the spot size is kept small. Lexan and Lucite are the desirable range shifter materials when the scattering power is kept similar to water.