An improvement in IMRT QA results and beam matching in linacs using statistical process control

The purpose of this study is to apply the principles of statistical process control (SPC) in the context of patient specific intensity‐modulated radiation therapy (IMRT) QA to set clinic‐specific action limits and evaluate the impact of changes to the multileaf collimator (MLC) calibrations on IMRT QA results. Ten months of IMRT QA data with 247 patient QAs collected on three beam‐matched linacs were retrospectively analyzed with a focus on the gamma pass rate (GPR) and the average ratio between the measured and planned doses. Initial control charts and action limits were calculated. Based on this data, changes were made to the leaf gap parameter for the MLCs to improve the consistency between linacs. This leaf gap parameter is tested monthly using a MLC sweep test. A follow‐up dataset with 424 unique QAs were used to evaluate the impact of the leaf gap parameter change. The initial data average GPR was 98.6% with an SPC action limit of 93.7%. The average ratio of doses was 1.003, with an upper action limit of 1.017 and a lower action limit of 0.989. The sweep test results for the linacs were ‐1.8%,0%, and +1.2% from nominal. After the adjustment of the leaf gap parameter, all sweep test results were within 0.4% of nominal. Subsequently, the average GPR was 99.4% with an SPC action limit of 97.3%. The average ratio of doses was 0.997 with an upper action limit of 1.011 and a lower action limit of 0.981. Applying the principles of SPC to IMRT QA allowed small differences between closely matched linacs to be identified and reduced. Ongoing analysis will monitor the process and be used to refine the clinical action limits for IMRT QA. PACS number: 87.55.Qr


I. InTRoducTIon
Statistical process control (SPC) has been used in industry since the early 1920s to reduce waste and increase the early detection and prevention capabilities of quality control (QC) and quality assurance (QA) systems. (1) Recently medical physicists have begun to apply the principles of SPC to all facets of the radiation oncology clinical practice. (2) Specifically, SPC can be used to identify out-of-control processes and improve a clinic's intensity-modulated radiation therapy (IMRT) QA program, linac QC program, and overall patient safety. (3)(4)(5)(6)(7)(8) Using these methods, the medical physicist can create clinic specific QA and QC action limits that are based on realworld data, as opposed to more generic national guidance. (3)(4)(5)(6)(7) In our clinical practice, we use three beam-matched linacs to treat IMRT patients. Patients are allowed to switch machines with little restriction. An analysis of IMRT performance outside the setting of daily, monthly, or annual QA had not been performed. This fact became a major motivator to study the three linacs in detail. The ability of SPC to identify out-of-control processes made it an excellent tool to verify if the three linacs were performing in a similar manner when delivering IMRT treatments. In this study we examined the initial application of SPC to our IMRT QA program, the effect of changes made to the QA program, and the longterm stability of the changes.

II. MATERIALS And METHodS
We retrospectively analyzed ten months of IMRT QA data with 247 patient-specific QAs collected on three beam-matched Varian Linear Accelerators (linacs) (21Ex and iX models) with Millennium 120 multileaf collimators (MLC) (Varian Medical Systems, Palo Alto, CA). No restrictions were put on beam energy; therefore, both of our clinically used photon energies, 6 MV and 18 MV, are represented in the pre-and postadjustment datasets. We focused on the gamma pass rate (GPR) and the average ratio between the measured and planned doses. All QAs were measured as true composites with a Sun Nuclear MapCHECK 2 (Sun Nuclear Corporation, Melbourne, FL) using the MapPHAN add-on (i.e., all fields were delivered at the planned gantry angles for fixed gantry plans or with the prescribed arcs), with the MapCHECK in either the coronal or sagittal plane. In order to deal with the angular dependence of the MapCHECK diodes, for static gantry IMRT we offset any beams entering parallel to the diode plane by at least 5° in the QA plan and measurement. We make no adjustments for rotating gantry IMRT; our tests with rotating open fields indicate that the perturbation is small. The GPR was computed by comparing the measured to calculated dose planes with 3%, 3 mm criteria. Additionally, the dose threshold was set at 10% and the Van Dyk difference, along with the measurement uncertainty capability of the Sun Nuclear patient software, was used. The calculated dose plane was created by mapping the patient plan onto a CT of the MapCHECK 2 with Hounsfield units redefined so as to appropriately model the response. This was verified by altering the Hounsfield units until the equivalent path of the MapCHECK 2 and MapPHAN system, as reported by Eclipse (Varian Medical Systems), equaled 5 cm. Doses were calculated with the Varian Eclipse AAA algorithm. The ratio between the measured and planned doses was calculated by averaging the results from the MapCHECK software's histogram of all data points within the 10% threshold. The standard deviation of that distribution has not been used as a quality measure because that variation is captured via the gamma distribution. As part of each IMRT QA, the MapCHECK response is adjusted to match the prediction for a standard open field, thus reducing variation caused by daily output fluctuation or detector response.
Initial control charts and action limits were calculated based on methods published by Breen et al. (3) The first 25 instances of a controlled system are used to calculate the upper control limit (UCL) and the lower control limit (LCL).
MR is defined as the average of the moving range and xis the mean of the data. This process was first applied retrospectively to the 247 data points obtained from IMRT QAs on all three accelerators to see if the process appeared to be in control and to determine the degree of variability.
Note that we treat the accelerators as matched and move patients between them freely, so we also do the IMRT QA test on any of the available machines. The justification for doing so depends on careful quality assurance of the MLC calibration, which we test both daily and monthly. The monthly test is performed along with the monthly output checks. A Farmer-type chamber is placed at 10 cm depth in water and the reading for a 10 × 10 cm 2 field given 100 MU is obtained. Then another reading is taken with a dynamic MLC motion that sweeps a 0.5 cm gap over the 10 cm span. The ratio of the two readings depends sensitively on the gap width, as shown in Fig. 1, and this can be adjusted in the MLC control software by changing the leaf gap parameter.
In order to investigate possible reasons for the observed variation in the initial dataset, the data were then analyzed for each accelerator separately to test whether there were systematic differences attributable to the machine, and if so, if there were concomitant variations in the monthly tests of MLC calibration that could be reduced and thereby reduce the variability in the IMRT QA results.
A follow-up dataset with 424 patient specific QAs was used to evaluate the effect of the MLC calibration tolerance change and the long-term stability of the IMRT QA program. The follow-up dataset has been analyzed in an identical manner as stated above.

III. RESuLTS
The results for the initial 247 IMRT QAs, taken prior to adjusting the tolerance of the MLC calibration test, are shown for the GPR in Fig. 2 and for the average dose ratio in Fig. 3. The preadjustment average GPR was 98.6% with an LCL of 93.7% (note that the GPR is singlesided with a maximum value of 100%, so only the LCL is relevant). The preadjustment average ratio of doses was 1.003 with a UCL of 1.017 and a LCL of 0.989. Table 1 shows the results in terms of means and standard deviations for the combined dataset and for the three accelerators separately, along with their variation in sweep test ratio. The difference between the accelerators was statistically significant, with p-values ranging from 9.3 × 10 -6 to 3.9 × 10 -9 , and paralleled the difference in sweep test ratio. The p-value for the dose was calculated using a two-tailed unpaired Student's t-test, while the p-value for GPR was calculated using a one-tailed unpaired Student's t-test. Based on these findings, the tolerance of the monthly sweep test ratio was changed from 0.0730-0.0785 to 0.0739-0.0772, while the nominal target stayed constant at 0.0758. The initial sweep test tolerances were based on historical clinical experience. The new This was done to confirm that the effect of altering the leaf gap offset value has on the sweep test value is both predictable and consistent across multiple types of Varian accelerators. tolerances were determined by calculating the upper and lower control limits by the SPC method described above, using the monthly sweep test values after the adjustment occurred.
The IMRT QA results following the adjustment are shown for the GPR in Fig. 2 and for the average dose ratio in Fig. 3. Postadjustment, the average GPR was 99.4% with an LCL of 97.3%, and the average dose ratio was 0.997 with an UCL of 1.011 and a LCL of 0.981. The   The ratio between the measured and planned doses was calculated by averaging the results from the MapCHECK software's histogram of all data points within the 10% threshold. b To measure the sweep test, a Farmer-type chamber is placed at 10 cm depth in water and the reading for a 10 × 10 cm 2 field given 100 MU is obtained. This reading is ratioed with reading that is taken with a dynamic MLC motion that sweeps a 0.5 cm gap over the 10 cm span. The nominal sweep test value is 0.0758. differences in these results in relation to the preadjustment values are statistically significant with a GPR p-value of 3.16 × 10 -11 and an average dose ratio p-value of 0.0058. Table 2 shows the results for the combined and separate datasets following the adjustment. The difference between the three accelerators remains statistically significant, although the overall variability has been reduced. This strengthened our judgment that the choice of machine for QA or treatment is not clinically significant. The MLC calibration continues to be tested on a monthly basis with the sweep test, and went from a preadjustment maximum deviation of 3% to a postadjustment maximum of 1.7% from the nominal value of 0.0758. The maximum deviation for the sweep test result was calculated using a simple percent difference compared to the nominal value stated above.

IV. dIScuSSIon
The application of SPC to our IMRT QA program has shown many benefits, both directly and indirectly. The 0.8% rise in the GPR, along with the 3.6% increase of the LCL, implies a distinct increase in the consistency of the IMRT QA results. The lack of improvement in the dose ratio values implies that this parameter is insensitive to the changes made to the MLC calibration. This is supported by the data in Table 1 and Table 2. A shift in the sweep test ratio did not result in a change to the average dose ratio. Further investigation into what accelerator parameters affect the dose ratio needs to occur as it appears to be a more complex issue. The decrease in the maximum range of the MLC calibration across the three linacs was largely responsible for the 50% decrease in the standard deviation of the data. The MLC calibration is arguably the most important factor for consistent IMRT and VMAT performance among the three beam matched linacs. (8)(9)(10) As described by Rangel and Dunscombe (9) and Moiseenko et al., (10) relatively small MLC errors can cause dosimetrically meaningful changes. These findings underscore the need to monitor and control the MLC calibration. There are still QAs that violate the control limits on a semiregular basis. Approximately 5% of QAs fall outside of the control limits. This value has remained constant pre-and post-MLC calibration adjustment. Each has been flagged for individual investigation, but so far no other systematic reason for the outliers has been found. During the period of this investigation, the proportion of IMRT treatments delivered by rotational techniques has increased from 60.7% prior to the adjustment of the sweep test tolerance to 84.9% after. It is possible that some of the reduction in variation can be attributed to this change in planning and delivery; however, that has not been evaluated.
This study has not addressed the important question of what the tolerance for the gamma pass rate and dose ratio should be to declare an IMRT plan safe for delivery. However it has allowed us to reduce the variability and better identify those plans deserving further scrutiny. It is important to note that our clinical practice of performing a true composite measurement has not been altered. Therefore, the gains described and the decrease in variability shows The ratio between the measured and planned doses was calculated by averaging the results from the MapCHECK software's histogram of all data points within the 10% threshold. b To measure the sweep test a Farmer-type chamber is placed at 10 cm depth in water and the reading for a 10 × 10 cm 2 field given 100 MU is obtained. This reading is ratioed with reading that is taken with a dynamic MLC motion that sweeps a 0.5 cm gap over the 10 cm span. The nominal sweep test value is 0.0758. the importance of applying SPC to IMRT QA, even if the measurement system has inherent angular dependancies.

V. concLuSIonS
Applying the principles of SPC to IMRT QA allowed small differences between closely matched linacs to be identified and reduced. Analysis has shown that the MLC calibration test results have a direct effect on the overall quality of the IMRT QA program. Declaring linacs to be beam-matched requires ongoing testing of the beam shaping devices, along with the dosimetric and mechanical properties. By testing the MLC calibration and maintaining strict pass criteria, it is possible to reduce the variability in IMRT QA results and improve the quality and safety of the QA program.