BackgroundThe development of a dose-volume-histogram (DVH) estimation model for knowledge-based planning is very time-consuming and it could be inefficient if it was only used for similar upcoming cases as supposed. It is clinically desirable to explore and validate other potential applications for a configured model. This study tests the hypothesis that a supine volumetric modulated arc therapy (VMAT) model can optimize intensity modulated radiotherapy (IMRT) plans of other patient setup orientations.MethodsBased on RapidPlan, a DVH estimation model was trained using 81 supine VMAT rectal plans and validated on 10 similar cases to ensure the robustness of its designed purpose. Attempts were then made to apply the model to re-optimize the dynamic MLC-sequences of the duplicated IMRT plans from 30 historical patients (20 prone and 10 supine) that were treated with the same prescription as for the model (50.6 and 41.8 Gy to 95 % of PGTV and PTV simultaneously/22 fractions). The performance of knowledge-based re-optimization and the impact of setup orientations were evaluated dosimetrically.ResultsThe VMAT model validation on similar cases showed comparable target dose distribution and significantly improved organ sparing (by 10.77 ~ 18.65 %) than the original plans. IMRT plans of either setup can be re-optimized using the supine VMAT model, which significantly reduced the dose to the bladder (by 25.88 % from 33.85 ± 2.96 to 25.09 ± 1.32 Gy for D50 %; by 22.77 % from 33.99 ± 2.77 to 26.25 ± 1.22 Gy for mean dose) and femoral head (by 12.27 % from 15.65 ± 3.33 to 13.73 ± 1.43 Gy for D50 %; by 10.09 % from 16.26 ± 2.74 to 14.62 ± 1.10 Gy for mean dose), all P < 0.01. Although the dose homogeneity and PGTV conformity index (CI_PGTV) changed slightly (≤0.01), CI_PTV of IMRT plans was significantly increased (Δ = 0.17, P < 0.01) by the manually defined target-objectives in the VMAT optimizer. The semi-automated IMRT planning increased the global maximum dose and V107 % due to the missing of hot spot suppression by specific manual optimizing or fluence map editing.ConclusionsThe Varian RapidPlan model trained on a technique and orientation can be used for another. Knowledge-based planning improves organ sparing and quality consistency, yet the target-objectives defined for VMAT-optimizer should be readapted to IMRT planning, followed by manual hot spot processing.

RapidPlan, a commercial knowledge‐based optimizer, has been tested on head and neck, lung, esophageal, breast, liver, and prostate cancer patients. To appraise its performance on VMAT planning with simultaneous integrated boosting (SIB) for rectal cancer, this study configured a DVH (dose‐volume histogram) estimation model consisting 80 best‐effort manual cases of this type. Using the model‐ generated objectives, the MLC (multileaf collimator) sequences of other 70 clinically approved plans were reoptimized, while the remaining parameters, such as field geometry and photon energy, were maintained. Dosimetric outcomes were assessed by comparing homogeneity index (HI), conformal index (CI), hot spots (volumes receiving over 107% of the prescribed dose, normalV107%), mean dose and dose to the 50% volume of femoral head (normalDmean_FH and normalD50%_FH), and urinary bladder (normalDmean_UB and normalD50%_UB), and the mean DVH plotting. Paired samples t‐test or Wilcoxon signed‐rank test suggested that comparable CI were achieved by RapidPlan (0.99 ± 0.04 for PTVboost, and 1.03 ± 0.02 for PTV) and original plans (1.00 ± 0.05 for PTVboost and 1.03 ± 0.02 for PTV), respectively (p > 0.05). Slightly improved HI of planning target volume (PTVboost) and PTV were observed in the RapidPlan cases (0.05 ± 0.01 for PTVboost, and 0.26 ± 0.01 for PTV) than the original plans (0.06 ± 0.01 for PTVboost and 0.26 ± 0.01 for PTV), p < 0.05. More cases with positive V107% were found in the original (18 plans) than the RapidPlan group (none). RapidPlan significantly reduced the normalD50%_FH (by 1.53 Gy/9.86% from 15.52 ± 2.17 to 13.99 ± 1.16 Gy), normalDmean_FH (by 1.29 Gy/7.78% from 16.59±2.07 to 15.30±0.70 G), normalD50%_UB (by 4.93 Gy/17.50% from 28.17±3.07 to 23.24±2.13 Gy), and normalDmean_UB (by 3.94 Gy/13.43% from 29.34±2.34 to 25.40±1.36 Gy), respectively. The more concentrated distribution of RapidPlan data points indicated an enhanced consistency of plan quality.PACS number(s): 87.55.de; 87.55.dk

Purpose The implementation of radiomics and machine learning (ML) techniques on analyzing two‐dimensional gamma maps has been demonstrated superior to the conventional gamma analysis for error identification in intensity modulated radiotherapy (IMRT) quality assurance (QA). Recently, the Structural SIMilarity (SSIM) sub‐index maps were shown to be able to reveal the error types of the dose distributions. In this study, we aimed to apply radiomics analysis on SSIM sub‐index maps and develop ML models to classify delivery errors in patient‐specific dynamic IMRT QA. Methods Twenty‐one sliding‐window IMRT plans of 180 beams for three treatment sites were involved in this study. Four types of machine‐related errors of various magnitudes were simulated for each beam at each control point, including the monitor unit (MU) variations, same‐directional and opposite‐directional shifts of the multileaf collimators (MLCs) and random mispositioning of the MLCs. In the QA process, a total of 1620 portal dose (PD) images were acquired for the beams with and without errors. The predicted PD images of the original beams were set as references. To quantify the agreement between a measured PD image and the corresponding predicted PD image, four difference maps including three SSIM sub‐index maps, and one dose difference‐derived map were calculated. Then, radiomic features were extracted from the four difference maps of each measured PD image. We tested four typical classifiers including linear discriminant classifier (LDC), two supporting vector machine (SVM) classifiers, and random forest (RF) for this multiclass classification task. A nested cross‐validation scheme was used for model evaluations, where the SVM recursive feature elimination method was applied for feature selection. Finally, the performance of the ML model on identifying the error‐free and the erroneous cases was compared to that of the conventional gamma analysis. Results The statistics of the selected features showed that all of the difference maps and the feature categories made balanced contributions to solve this classification task. Best performance was achieved by the Linear‐SVM model with average overall classification accuracy of 0.86. Specifically, the average classification accuracies of the shift, opening, and the random errors were around 0.9. Moreover, ~80% of error‐free and MU errors were correctly classified. Using gamma analysis, the 3 mm/3% criterion was found insensitive to errors (sensitivity was only 0.33). Although the sensitivity to errors with the 2 mm/2% criterion increased to 0.79, still 8% worse than that of the ML model. Conclusions We proposed an ML‐based method for machine‐related error identification in patient‐specific dynamic IMRT QA, where radiomic analysis on SSIM sub‐index maps were used for feature extraction. With extensive validation to select the best features and classifiers, high accuracies in error classification were achieved. Compared with the conventional gamma threshold method, this approach has great potential in error...

PurposeTo test if a RapidPlan DVH estimation model and its training plans can be improved interactively through a closed‐loop evolution process.Methods and materialsEighty‐one manual plans (P0) that were used to configure an initial rectal RapidPlan model (M0) were reoptimized using M0 (closed‐loop), yielding 81 P1 plans. The 75 improved P1 (P1+) and the remaining 6 P0 were used to configure model M1. The 81 training plans were reoptimized again using M1, producing 23 P2 plans that were superior to both their P0 and P1 forms (P2+). Hence, the knowledge base of model M2 composed of 6 P0, 52 P1+, and 23 P2+. Models were tested dosimetrically on 30 VMAT validation cases (Pv) that were not used for training, yielding Pv(M0), Pv(M1), and Pv(M2) respectively. The 30 Pv were also optimized by M2_new as trained by the library of M2 and 30 Pv(M0).ResultsBased on comparable target dose coverage, the first closed‐loop reoptimization significantly (P < 0.01) reduced the 81 training plans’ mean dose to femoral head, urinary bladder, and small bowel by 2.65 Gy/15.63%, 2.06 Gy/8.11%, and 1.47 Gy/6.31% respectively, which were further reduced significantly (P < 0.01) in the second closed‐loop reoptimization by 0.04 Gy/0.28%, 0.18 Gy/0.77%, 0.22 Gy/1.01% respectively. However, open‐loop VMAT validations displayed more complex and intertwined plan quality changes: mean dose to urinary bladder and small bowel decreased monotonically using M1 (by 0.34 Gy/1.47%, 0.25 Gy/1.13%) and M2 (by 0.36 Gy/1.56%, 0.30 Gy/1.36%) than using M0. However, mean dose to femoral head increased by 0.81 Gy/6.64% (M1) and 0.91 Gy/7.46% (M2) than using M0. The overfitting problem was relieved by applying model M2_new.ConclusionsThe RapidPlan model and its constituent plans can improve each other interactively through a closed‐loop evolution process. Incorporating new patients into the original training library can improve the RapidPlan model and the upcoming plans interactively.

The unwanted radiation transmission through the multileaf collimators could be reduced by the jaw tracking technique which is commercially available on Varian TrueBeam accelerators. On the basis of identical plans, this study aims to investigate the dosimetric impact of jaw tracking on the volumetric‐modulated arc therapy (VMAT) plans. Using Eclipse treatment planning system (TPS), 40 jaw‐tracking VMAT plans with various tumor volumes and shapes were optimized. Fixed jaw plans were created by editing the jaw coordinates of the jaw‐tracking plans while other parameters were identical. The deliverability of this artificial modification was verified using COMPASS system via three‐dimentional gamma analysis between the measurement‐based reconstruction and the TPS‐calculated dose distribution. Dosimetric parameters of dose‐volume histogram (DVH) were compared to assess the improvement of dose sparing for organs at risk (OARs) in jaw‐tracking plans. COMPASS measurements demonstrated that over 96.9% of structure volumes achieved gamma values less than 1.00 at criteria of 3 mm/3%. The reduction magnitudes of maximum and mean dose to various OARs ranged between 0.06%∼6.76%false(0.04∼7.29 Gyfalse) and 0.09%∼7.81%false(0.02∼2.78 Gyfalse), respectively, using jaw tracking, agreeing with the disparities of radiological characteristics between MLC and jaws. Jaw tracking does not change the delivery efficiency and total monitor units. The dosimetric comparison of VMAT plans with and without jaw tracking confirms the physics hypotheses that reduced transmission through tracking jaws will reduce doses to OARs without sacrificing the target dose coverage because it is meant to be covered by radiation beams going through the opening.PACS number(s): 87.55.de, 87.55.dk