Ding, Xiaoning

Abstract Background In the current clinical workflow of radiation oncology, therapists manually summarize the physician‐issued Computed Tomography (CT) simulation orders to prepare patients' CT simulations. This process increases the workload, introduces variability in documentation quality, and is prone to human errors. Purpose This study aims to address these challenges using a large language model (LLM) to automate the generation of summaries from the CT simulation orders and evaluate its performance. Methods A total of 607 CT simulation orders were collected from the Aria database at our institution. A locally hosted LLaMa 3.1 405B model, accessed via the Application Programming Interface (API) service, was used to extract keywords from the CT simulation orders and generate summaries. The downloaded CT simulation orders were categorized into seven groups based on treatment modalities and disease sites. For each group, a customized instruction prompt was developed collaboratively with therapists to guide the LLaMa 3.1 405B model in generating summaries. The ground truth for the corresponding summaries was manually derived by carefully reviewing each CT simulation order and subsequently verified by therapists. The accuracy of the LLM‐generated summaries was evaluated by therapists using the ground truth. Results Over 98% of the LLM‐generated summaries aligned with the ground truth in terms of accuracy. Our evaluations showed an improved consistency in format and enhanced readability from the LLM‐generated summaries compared to the corresponding therapist‐generated summaries. This automated approach demonstrated consistent performance across all groups, regardless of treatment modality or disease site. Conclusions This study demonstrated the high precision and consistency of the LLaMa 3.1 405B model in extracting keywords and summarizing CT simulation orders, suggesting that LLMs have great potential to assist in this task, reduce the workload of CT simulation therapists and improve radiation oncology workflow efficiency.

AbstractPurposeTo measure diode sensitivity degradation (DSD) induced by cumulative proton dose delivered to a commercial daily quality assurance (QA) device.MethodsAt our institution, six Daily QA 3 (DQA3, Sun Nuclear Corporation, Melbourne, FL, USA) devices have been used for daily proton pencil beam scanning QA in four proton gantry rooms over a span of 4 years. DQA3 diode counts were cross‐calibrated using a homogenous field with a known dose of 1 Gy. The DSD rate (ΔR%/100 Gy) was calculated using linear regression on time‐series plots of diode counts and an estimate of cumulative dose per year based on the cross‐calibration. The effect of DSD on daily QA spot position measurements was quantified by converting DSD to baseline spot position shift.ResultsThe average dose delivered to the four inner DQA3 diodes was 104 ± 5 Gy/year, and the rate of DSD was ‐5.1% ± 1.0/100 Gy with the exception of one DQA3 device that had a significantly higher rate of DSD (–12%/100 Gy). The R2s of the linear fit to time‐series plots were between 0.92 and 0.98. The DSD rates were not constant but decreases with accumulated doses. The four center diodes, which received 40% of the cumulative dose received by inner diodes, had a DSD rate of –7.2% ± 0.9/100 Gy. For our daily QA program, 1 year of DSD was equivalent to a 0.2 mm shift in spot position.ConclusionsThe DSD rate of DQA3 diodes determined by long‐term proton daily QA data was about –5%/100 Gy, which is more than 10 times greater than the reported DSD rate from photon irradiation. DQA3 diodes may be used for daily proton QA programs, provided that they are recalibrated at an appropriate frequency that should be determined specifically for different daily QA programs.

AbstractPurposeTo validate breast tissue expander metallic port (MP) models in a commercial treatment planning system (TPS) in proton pencil beam scanning (PBS) treatments for breast cancer patients with breast tissue expanders.Methods and materialsThree types of MPs taken out of a Mentor CPX4, a Natrelle 133, and a PMT Integra breast tissue expanders and a 650 cc saline filled Mentor CPX4 expander were placed on top of acrylic slabs, and scanned using a Siemens Somatom Definition AS Open RT CT scanner. Structure templates for each of the MPs were designed within Eclipse TPS. The CT numbers for the metallic parts were overridden to reflect measured or calculated relative proton stopping powers (RPSPs). Mock targets were contoured in acrylic to represent postmastectomy chest‐wall radiation therapy (PMRT) targets. Plans with different beam incident angles were optimized using the Eclipse TPS to deliver uniform prescription dose to the target using Hitachi Probeat‐V PBS beams. Eclipse calculated doses and an in‐house Monte Carlo (MC) code calculated doses were compared to the measured Gafchromic EBT3 film doses in acrylic.ResultsTPS/MC and film dose comparison results showed that (1) 3%/2 mm/10% threshold Gamma pass rates were better than 90.8% in the acrylic target region for all plans; (2) comparing TPS and film doses for the individual beam plans in the MP dose shadow areas, the area with dose difference above 5% ([ΔA] 5%) ranged from 1.1 to 5.0 cm2, and the maximum dose difference ([ΔD] 0.01 cm2) ranged from 12.5% to 25.0%; (3) comparing MC and film doses for the individual beam plans in the MP dose shadow areas, the (ΔA) 5% varied from 1.1 to 2.9 cm2 and (ΔD) 0.01 cm2 varied from 8.5% to 24.2%; (4) for a plan composed of three individual beams treating through the Mentor CPX4 expander, the TPS (ΔA) 5% was less than 0.13 cm2, and the (ΔD) 0.01 cm2 was less than 6% in the MP dose shadow areas.ConclusionsIt is feasible to treat patients with tissue expanders using multiple PBS beams using a structure template with CT number overridden to represent the measured/calculated RPSP for MPs for PBS treatment planning. MC dose was more accurate than analytical dose in the areas with high dose gradient caused by the density heterogeneity of the breast tissue expander MPs.

Abstract Purpose The multiple energy extraction (MEE) delivery technique for synchrotron‐based proton delivery systems reduces beam delivery time by decelerating the beam multiple times during one accelerator spill, but this might cause additional plan quality degradation due to intrafractional motion. We seek to determine whether MEE causes significantly different plan quality degradation compared to single energy extraction (SEE) for lung cancer treatments due to the interplay effect. Methods Ten lung cancer patients treated with IMPT at our institution were nonrandomly sampled based on a representative range of tumor motion amplitudes, tumor volumes, and respiratory periods. Dose‐volume histogram (DVH) indices from single‐fraction SEE and MEE four‐dimensional (4D) dynamic dose distributions were compared using the Wilcoxon signed‐rank test. Distributions of monitor units (MU) to breathing phases were investigated for features associated with plan quality degradation. SEE and MEE DVH indices were compared in fractionated deliveries of the worst‐case patient treatment scenario to evaluate the impact of fractionation. Results There were no clinically significant differences in target mean dose, target dose conformity, or dose to organs‐at‐risk between SEE and MEE in single‐fraction delivery. Three patients had significantly worse dose homogeneity with MEE compared to SEE (single‐fraction mean D 5% ‐D 95% increased by up to 9.6% of prescription dose), and plots of MU distribution to breathing phases showed synchronization patterns with MEE but not SEE. However, after 30 fractions the patient in the worst‐case scenario had clinically acceptable target dose homogeneity and coverage with MEE (mean D 5% –D 95% increased by 1% compared to SEE). Conclusions For some patients with breathing periods close to the mean spill duration, MEE resulted in significantly worse single‐fraction target dose homogeneity compared to SEE due to the interplay effect. However, this was mitigated by fractionation, and target dose homogeneity and coverage were clinically acceptable after 30 fractions with MEE.

AbstractPurposeTo demonstrate that variation in chemical composition has a negligible effect on the mapping curve from relative electron density (RED) to proton stopping power ratio (SPR), and to establish the theoretical framework of using Megavoltage (MV) computed tomography (CT), instead of kilovoltage (kV) dual energy CT, to accurately estimate proton SPR.MethodsA simulation study was performed to evaluate the effect of chemical composition variation on kVCT number and proton SPR. The simulation study involved both reference and simulated human tissues. The reference human tissues, together with their physical densities and chemical compositions, came from the ICRP publication 23. The simulated human tissues were created from the reference human tissues assuming that elemental percentage weight followed a Gaussian distribution. For all tissues, kVCT number and proton SPR were obtained through (a) theoretical calculation from tissue’s physical density and chemical composition which served as the ground truth, and (b) estimation from RED using the calibration curves established from the stoichiometric method. Deviations of the estimated values from the calculated values were quantified as errors in using RED to estimate kVCT number and proton SPR.ResultsGiven a chemical composition variation of 5% (1σ) of the nominal percentage weights, the total estimation error of using RED to estimate kVCT number was 0.34%, 0.62%, and 0.77% and the total estimation error of using RED to estimate proton SPR was 0.30%, 0.22%, and 0.16% for fat tissues, non‐fat soft tissues and bone tissues, respectively.ConclusionChemical composition had a negligible effect on the method of using RED to determine proton SPR. RED itself is sufficient to accurately determine proton SPR. MVCT number maintains a superb linear relationship with RED because it is highly dominated by Compton scattering. Therefore, MVCT has great potential in reducing the proton range uncertainty.

PurposeTo commission an open source Monte Carlo (MC) dose engine, “MCsquare” for a synchrotron‐based proton machine, integrate it into our in‐house C++‐based I/O user interface and our web‐based software platform, expand its functionalities, and improve calculation efficiency for intensity‐modulated proton therapy (IMPT).MethodsWe commissioned MCsquare using a double Gaussian beam model based on in‐air lateral profiles, integrated depth dose of 97 beam energies, and measurements of various spread‐out Bragg peaks (SOBPs). Then we integrated MCsquare into our C++‐based dose calculation code and web‐based second check platform “DOSeCHECK.” We validated the commissioned MCsquare based on 12 different patient geometries and compared the dose calculation with a well‐benchmarked GPU‐accelerated MC (gMC) dose engine. We further improved the MCsquare efficiency by employing the computed tomography (CT) resampling approach. We also expanded its functionality by adding a linear energy transfer (LET)‐related model‐dependent biological dose calculation.ResultsDifferences between MCsquare calculations and SOBP measurements were <2.5% (<1.5% for ~85% of measurements) in water. The dose distributions calculated using MCsquare agreed well with the results calculated using gMC in patient geometries. The average 3D gamma analysis (2%/2 mm) passing rates comparing MCsquare and gMC calculations in the 12 patient geometries were 98.0 ± 1.0%. The computation time to calculate one IMPT plan in patients’ geometries using an inexpensive CPU workstation (Intel Xeon E5‐2680 2.50 GHz) was 2.3 ± 1.8 min after the variable resolution technique was adopted. All calculations except for one craniospinal patient were finished within 3.5 min.ConclusionsMCsquare was successfully commissioned for a synchrotron‐based proton beam therapy delivery system and integrated into our web‐based second check platform. After adopting CT resampling and implementing LET model‐dependent biological dose calculation capabilities, MCsquare will be sufficiently efficient and powerful to achieve Monte Carlo‐based and LET‐guided robust optimization in IMPT, which will be done in the future studies.

PurposeThe dose‐averaged linear energy transfer (LETd) for intensity‐modulated proton therapy (IMPT) calculated by one‐dimensional (1D) analytical models deviates from more accurate but time‐consuming Monte Carlo (MC) simulations. We developed a fast hybrid three‐dimensional (3D) analytical LETd calculation that is more accurate than 1D analytical model.MethodsWe used the Geant4 MC code to generate 3D LETd distributions of monoenergetic proton beams in water for all energies and used a customized error function to fit the LETd lateral profiles at various depths to the MC simulation. The 3D LETd calculation kernel was a lookup table of these fitted coefficients, and LETd was determined directly from spot energies and voxel coordinates during analytical dose calculations. We validated our new method by comparing the calculated LETd distributions to MC results using 3D Gamma index analysis with 3%/2 mm criteria in 12 patient geometries. The significance of the improvement in Gamma index analysis passing rates over the 1D analytical model was determined using the Wilcoxon rank‐sum test.ResultsThe passing rate of 3D Gamma analysis comparing LETd distributions from the hybrid 3D method and the 1D method to MC simulations was significantly improved from 94.0% ± 2.5% to 98.0% ± 1.0% (P = 0.0003). The typical time to calculate dose and LETd simultaneously using an Intel Xeon E5‐2680 2.50 GHz workstation was approximately 2.5 min.ConclusionsOur new method significantly improved the LETd calculation accuracy compared to the 1D method while maintaining significantly shorter calculation time even comparing with the GPU‐based fast MC code.

Purpose: To describe and validate the dose calculation algorithm of an independent second-dose check software for spot scanning proton delivery systems with full width at half maximum between 5 and 14 mm and with a negligible spray component. Methods: The analytical dose engine of our independent second-dose check software employs an altered pencil beam algorithm with 3 lateral Gaussian components. It was commissioned using Geant4 and validated by comparison to point dose measurements at several depths within spread-out Bragg peaks of varying ranges, modulations, and field sizes. Water equivalent distance was used to compensate for inhomogeneous geometry. Twelve patients representing different disease sites were selected for validation. Dose calculation results in water were compared to a fast Monte Carlo code and ionization chamber array measurements using dose planes and dose profiles as well as 2-dimensional–3-dimensional and 3-dimensional–3-dimensional γ-index analysis. Results in patient geometry were compared to Monte Carlo simulation using dose–volume histogram indices, 3-dimensional–3-dimensional γ-index analysis, and inpatient dose profiles. Results: Dose engine model parameters were tuned to achieve 1.5% agreement with measured point doses. The in-water γ-index passing rates for the 12 patients using 3%/2 mm criteria were 99.5% ± 0.5% compared to Monte Carlo. The average inpatient γ-index analysis passing rate compared to Monte Carlo was 95.8% ± 2.9%. The average difference in mean dose to the clinical target volume between the dose engine and Monte Carlo was −0.4% ± 1.0%. For a typical plan, dose calculation time was 2 minutes on an inexpensive workstation. Conclusions: Following our commissioning process, the analytical dose engine was validated for all treatment sites except for the lung or for calculating dose–volume histogram indices involving point doses or critical structures immediately distal to target volumes. Monte Carlo simulations are recommended for these scenarios.

PurposeTo clinically implement and comprehensively evaluate two independent methods for beam monitor calibration of scanning proton beam.MethodsSeven proton energies that represent the lowest to highest energy proton beams were selected. Single energy layer circular fields of diameter 15 cm with 2.5 mm spot spacing and 10 times of repainting (FS15cm) were designed for all energies. The effective measurement points of Bragg peak chamber (BPC), advanced Markus chamber (AMC) and farmer chamber (FC) were all aligned to 2 cm depth in water using SSD setup. The BPC and AMC were cross‐calibrated with farmer chamber (FC) using the field FS15cm. In order to evaluate BPC's lateral response uniformity, a collimated narrow proton beam (5.8 mm diameter) was delivered to the active area and edge of the BPC. The dose area product (DAP) was measured using two methods by two BPCs, one AMC and one FC. For method 1, a single spot proton beam was delivered to the geometric center of the BPC. For method 2, the fields FS15cm were delivered to FC and AMC, respectively. Accumulated charges by these chambers were converted to DAPs, and the quantitative difference of DAPs between both methods was calculated. The causes of the uncertainties were discussed, and the advantages of the two methods were compared.ResultsThe two BPCs showed different lateral response uniformity. BPC1 has a uniform response from the center up to a radius of 3.5 cm. BPC2 has a uniform response only to 2 cm and the response dropped 1% to 2% at 3.5 cm from center. BPC2 also has significant over‐response compared to BPC1. A 2.2% systematic error would be transferred to DAP if the over‐response from BPC2 was not considered. The DAPs measured by method 1 with two BPCs and by method 2 with FC and AMC were consistent to 0.5%. The major uncertainty component of method 1 is from the cross‐calibration of the BPC.ConclusionsThe two independent methods for DAP were shown to give consistent results, given the sources of uncertainties were carefully addressed in the measurements. Direct measurement of DAP with BPC is very efficient, but it may be subject to more than 2% systematic error if the BPC lateral response is not carefully evaluated.

PurposeApproximate dose calculation methods were used in the nominal dose distribution and the perturbed dose distributions due to uncertainties in a commercial treatment planning system (CTPS) for robust optimization in intensity‐modulated proton therapy (IMPT). We aimed to investigate whether the approximations influence plan quality, robustness, and interplay effect of the resulting IMPT plans for the treatment of locally advanced lung cancer patients.Materials and methodsTen consecutively treated locally advanced nonsmall cell lung cancer (NSCLC) patients were selected. Two IMPT plans were created for each patient using our in‐house developed TPS, named “Solo,” and also the CTPS, EclipseTM (Varian Medical Systems, Palo Alto, CA, USA), respectively. The plans were designed to deliver prescription doses to internal target volumes (ITV) drawn by a physician on averaged four‐dimensional computed tomography (4D‐CT). Solo plans were imported back to CTPS, and recalculated in CTPS for fair comparison. Both plans were further verified for each patient by recalculating doses in the inhalation and exhalation phases to ensure that all plans met clinical requirements. Plan robustness was quantified on all phases using dose‐volume‐histograms (DVH) indices in the worst‐case scenario. The interplay effect was evaluated for every plan using an in‐house developed software, which randomized starting phases of each field per fraction and accumulated dose in the exhalation phase based on the patient's breathing motion pattern and the proton spot delivery in a time‐dependent fashion. DVH indices were compared using Wilcoxon rank‐sum test.ResultsCompared to the plans generated using CTPS on the averaged CT, Solo plans had significantly better target dose coverage and homogeneity (normalized by the prescription dose) in the worst‐case scenario [ITV D95%: 98.04% vs 96.28%, Solo vs CTPS, P = 0.020; ITV D5%–D95%: 7.20% vs 9.03%, P = 0.049] while all DVH indices were comparable in the nominal scenario. On the inhalation phase, Solo plans had better target dose coverage and cord Dmax in the nominal scenario [ITV D95%: 99.36% vs 98.45%, Solo vs CTPS, P = 0.014; cord Dmax: 20.07 vs 23.71 Gy(RBE), P = 0.027] with better target coverage and cord Dmax in the worst‐case scenario [ITV D95%: 97.89% vs 96.47%, Solo vs CTPS, P = 0.037; cord Dmax: 24.57 vs 28.14 Gy(RBE), P = 0.037]. On the exhalation phase, similar phenomena were observed in the nominal scenario [ITV D95%: 99.63% vs 98.87%, Solo vs CTPS, P = 0.037; cord Dmax: 19.67 vs 23.66 Gy(RBE), P = 0.039] and in the worst‐case scenario [ITV D95%: 98.20% vs 96.74%, Solo vs CTPS, P = 0.027; cord Dmax: 23.47 vs 27.93 Gy(RBE), P = 0.027]. In terms of interplay effect, plans generated by Solo had significantly better target dose coverage and homogeneity, less hot spots, and lower esophageal Dmean, and cord Dmax [ITV D95%: 101.81% vs 98.68%, Solo vs CTPS, P = 0.002; ITV D5%–D95%: 2.94% vs 7.51%, P = 0.002; cord Dmax: 18.87 vs 22.29 Gy(RBE), P = 0.014].ConclusionsSolo‐generated IMPT plans provide improved cord sparing, better target robustness in all considered phases, and reduced interplay effect compared with CTPS. Consequently, the approximation methods currently used in commercial TPS programs may have space for improvement in generating optimal IMPT plans for patient cases with locally advanced lung cancer.