This work applied statistical process control to establish the control limits of the % gamma pass of patient-specific intensity modulated radiotherapy (IMRT) and volumetric modulated arc therapy (VMAT) quality assurance (QA), and to evaluate the efficiency of the QA process by using the process capability index (Cpml). A total of 278 IMRT QA plans in nasopharyngeal carcinoma were measured with MapCHECK, while 159 VMAT QA plans were undertaken with ArcCHECK. Six megavolts with nine fields were used for the IMRT plan and 2.5 arcs were used to generate the VMAT plans. The gamma (3%/3 mm) criteria were used to evaluate the QA plans. The % gamma passes were plotted on a control chart. The first 50 data points were employed to calculate the control limits. The Cpml was calculated to evaluate the capability of the IMRT/VMAT QA process. The results showed higher systematic errors in IMRT QA than VMAT QA due to the more complicated setup used in IMRT QA. The variation of random errors was also larger in IMRT QA than VMAT QA because the VMAT plan has more continuity of dose distribution. The average % gamma pass was 93.7% ± 3.7% for IMRT and 96.7% ± 2.2% for VMAT. The Cpml value of IMRT QA was 1.60 and VMAT QA was 1.99, which implied that the VMAT QA process was more accurate than the IMRT QA process. Our lower control limit for % gamma pass of IMRT is 85.0%, while the limit for VMAT is 90%. Both the IMRT and VMAT QA processes are good quality because Cpml values are higher than 1.0.

The percentage depth dose in the build-up region and the surface dose for the 6-MV photon beam from a Varian Clinac 23EX medical linear accelerator was investigated for square field sizes of 5 × 5, 10 × 10, 15 × 15 and 20 × 20 cm2using the EGS4nrc Monte Carlo (MC) simulation package. The depth dose was found to change rapidly in the build-up region, and the percentage surface dose increased proportionally with the field size from approximately 10% to 30%. The measurements were also taken using four common detectors: TLD chips, PFD dosimeter, parallel-plate and cylindrical ionization chamber, and compared with MC simulated data, which served as the gold standard in our study. The surface doses obtained from each detector were derived from the extrapolation of the measured depth doses near the surface and were all found to be higher than that of the MC simulation. The lowest and highest over-responses in the surface dose measurement were found with the TLD chip and the CC13 cylindrical ionization chamber, respectively. Increasing the field size increased the percentage surface dose almost linearly in the various dosimeters and also in the MC simulation. Interestingly, the use of the CC13 ionization chamber eliminates the high gradient feature of the depth dose near the surface. The correction factors for the measured surface dose from each dosimeter for square field sizes of between 5 × 5 and 20 × 20 cm2are introduced.

Shewhart control charts have previously been suggested as a process control tool for use in routine linear accelerator (linac) output verifications. However, a comprehensive approach to process control has not been investigated for linac output verifications. The purpose of this work is to investigate a comprehensive process control approach to linac output constancy quality assurance (QA). The RBA‐3 dose constancy check was used to verify outputs of photon beams and electron beams delivered by a Varian Clinac 21EX linac. The data were collected during 2009 to 2010. Shewhart‐type control charts, exponentially weighted moving average (EWMA) charts, and capability indices were applied to these processes. The Shewhart‐type individuals chart (X‐chart) was used and the number of data points used to calculate the control limits was varied. The parameters tested for the EWMA charts (smoothing parameter (λ) and the control limit width (L)) were λ=0.05, normalL=2.492; λ=0.10, normalL=2.703; and λ=0.20, normalL=2.860, as well as the number of points used to estimate the initial process mean and variation. Lastly, the number of in‐control data points used to determine process capability (Cp) and acceptability (Cpk) were investigated, comparing the first in‐control run to the longest in‐control run of the process data. Cnormalp and Cpk values greater than 1.0 were considered acceptable. The 95% confidence intervals were reported. The X‐charts detected systematic errors (e.g., device setup errors). In‐control run lengths on the X‐charts varied from 5 to 30 output measurements (about one to seven months). EWMA charts showed in‐control runs ranging from 9 to 33 output measurements (about two to eight months). The Cnormalp and Cpk ratios are higher than 1.0 for all energies, except 12 and 20 MeV. However, 10 MV and 6, 9, and 16 MeV were in question when considering the 95% confidence limits. The X‐chart should be calculated using 8–12 data points. For EWMA chart, using 4 data points is sufficient to calculate the initial mean and variance of the process. The EWMA limits should be calculated with λ=0.10, normalL=2.703. At least 25–30 in‐control data points should be used to calculate the Cnormalp and Cpk indices.PACS number: 89

The purpose of this study was to develop a predictive model for patient‐specific VMAT QA results using multileaf collimator (MLC) effect and texture analysis. The MLC speed, acceleration and texture analysis features were extracted from 106 VMAT plans as predictors. Gamma passing rate (GPR) was collected as a response class with gamma criteria of 2%/2 mm and 3%/2 mm. The model was trained using two machine learning methods: AdaBoost classification and bagged regression trees model. GPR was classified into the “PASS” and “FAIL” for the classification model using the institutional warning level. The accuracy of the model was assessed using sensitivity and specificity. In addition, the accuracy of the regression model was determined using the difference between predicted and measured GPR. For the AdaBoost classification model, the sensitivity/specificity was 94.12%/100% and 63.63%/53.13% at gamma criteria of 2%/2 mm and 3%/2 mm, respectively. For the bagged regression trees model, the sensitivity/specificity was 94.12%/91.89% and 61.18%/68.75% at gamma criteria of 2%/2 mm and 3%/2 mm, respectively. The root mean square error (RMSE) of difference between predicted and measured GPR was found at 2.44 and 1.22 for gamma criteria of 2%/2 mm and 3%/2 mm, respectively. The promising result was found at tighter gamma criteria 2%/2 mm with 94.12% sensitivity (both bagged regression trees and AdaBoost classification model) and 100% specificity (AdaBoost classification model).

No abstract