Quality Assurance of Radiotherapy Dose Calculations

Dosimetric Testing

Dosimetric data can be in the form of point doses, 2D isodoses, 3D isodoses, dose/volume information, or activity. To test the accuracy of the calculations from a system, three strategies can be employed:

• 1.

  Comparison of calculated doses to measurements

• 2.

  Comparison of calculated doses to benchmark data

• 3.

  Comparison of calculated doses from one system to another system that has already been validated

A full commissioning of a system will usually involve using at least two of these strategies.

Another factor that must be evaluated is the dose calculation grid resolution. A system may give clinically different results for a 4 mm grid spacing compared to a 2 mm grid. It is very important to recognize when to use smaller or larger grids. For example, some studies recommend a 2.5 mm or smaller grids for intensity-modulated radiation therapy (IMRT) calculations, and the grid spacing should be proportional to the multi-leaf collimator (MLC) leaf width, meaning that a micro-MLC will need 2 mm or smaller grid spacing. Other special considerations include the use of Monte Carlo–based treatment planning to account for tissue inhomogeneities, the dose effects of couch tops, and biological models.

The user must be aware of how the TPS reports dose to heterogeneities. The system may calculate dose to the medium or convert to dose to water by using stopping power ratios. These two methods will lead to different results. The relative merits have been discussed in the literature but no standard has yet emerged. This issue is of particular concern for MC algorithms. See AAPM TG-105, Issues Associated with Clinical Implementation of Monte Carlo–Based Photon and Electron External Beam Treatment Planning, for details. Monte Carlo calculations have long been used to accurately calculate doses, particularly in heterogeneous tissues where electron transport issues are not properly handled by model-based calculations. The long calculation times for Monte Carlo algorithms have prevented their general use in the clinic. With the advent of faster computing hardware and optimized codes, they are now entering the mainstream. The issues of statistical uncertainty, the ability to account for the exact geometry of the accelerator treatment head, and other features are unique to Monte Carlo algorithms and must be well understood for accurate implementation. For example, because of the statistical uncertainty of the dose to any voxel, normalization to point doses can lead to unexpected results. This is especially relevant for electron Monte Carlo, where treatment dose is often prescribed to d _max . For dose prescriptions to an isodose line, a Monte Carlo calculation uncertainty of 2% at prescription isodose has evolved as a clinical standard. It should be noted, though, that uncertainties of dose calculation are higher for organs at risks (OARs) receiving lower doses than the prescription dose. A Monte Carlo treatment planning system should be able to display the uncertainty of dose calculation on the treatment planning scan to allow the clinicians to thoroughly assess the dose distribution.

AAPM TG-176, Dosimetric Effects Caused by Couch Tops and Immobilization Devices, discusses how the treatment couch and immobilization devices can cause increased skin dose, reduced tumor dose, and a change in the dose distribution if not properly addressed. In particular, beam entrance through high density components should be avoided. Attenuation through carbon fiber couch tops can range from 2% for normally incident beams to 6% for highly oblique beams at low energies. In addition, denser sections can have up to 17% attenuation. The attenuation will increase if the couch top is paired with other immobilization devices. The skin dose can reach 100% of the maximum dose. Simple attenuation corrections can lead to inaccurate dose calculation, and the explicit inclusion of the devices by the TPS is recommended.

AAPM TG-166, The Use and QA of Biologically Related Models for Treatment Planning, discusses four issues with biological models:

• 1.

  Strategies, limitations, conditions, and cautions for using biological models

• 2.

  The practical use of the most commonly used biological models: equivalent uniform dose (EUD), normal tissue complication probability (NTCP), tumor control probability (TCP), and probability of complication free tumor control (P ₊ )

• 3.

  Desirable features and future directions of biological models

• 4.

  General guidelines and methodology for acceptance testing, commissioning, and ongoing QA of biological models

Because the use of biological models represents a paradigm shift, they must be implemented carefully to avoid a dangerous misapplication of the models. The use of biological models can have advantages in both optimization and plan evaluation. This is because if the model is accurate the results are directly correlated with outcomes and can more reliably rank plans for a given patient. More detail on biological models is given in Chapter 14 .

Comparison of Calculated Doses to Measurements

A forthcoming Medical Physics Practice Guideline (MPPG) #5 from AAPM on Commissioning and QA of TPS in external beam therapy will have a summary of the minimum requirements for dosimetric testing (AAPM TG-244). Nondosimetric testing is not part of the scope of the planned document.

When commissioning a TPS there are two types of measurements taken: beam data needed to characterize the beam model and data needed to validate the model. For linear accelerators, this is the subject of AAPM TG-106 on accelerator beam data commissioning equipment and procedure. The characterization set varies by vendor but includes items such as output factors, wedge transmission factors, depth dose curves, and profiles. These are typically measured in a water phantom. Specifics related to these measurements can be found in Chapter 2 . The user must be careful to perform the measurements under the conditions described by the vendor. Particular attention must be paid to the definition of the normalization condition and the definition of wedge transmission. For instance, some systems normalize to 90 cm source-to-surface distance (SSD) and 10 cm depth while others require 100 cm SSD. In addition, some systems require wedge transmission relative to a 10 × 10 cm field while others require a ratio of wedge to open field for each field size.

As discussed in Chapter 2 , the measurement of surface dose must be done using the appropriate detector (extrapolation or parallel plate chamber). This is typically an area of weakness for many dose calculation algorithms and should be carefully evaluated. If the system cannot accurately calculate surface dose, more frequent patient-specific measurements may be required to confirm the skin dose.

Out-of-field doses are important for characterizing the risk for situations such as a pregnant patient and patients with implanted electronic devices. AAPM TG-158, Measurements and Calculations of Doses Outside the Treatment Volume from External Beam Radiation Therapy, is developing recommendations in this area. The accuracy of out-of-field dose measurements should be checked using an anthropomorphic phantom. In recent years, a few clinical studies have emerged emphasizing the importance of assessing integral dose and dose to distant organs to avoid late secondary cancers. Studies have shown that some systems may have difficulty in calculating these doses accurately. At distances close to the irradiated volume, internal scatter is the largest component of the dose, whereas at larger distances machine scatter and leakage dominate. For 3D conformal systems internal scatter contributes up to 70% of the dose and dominates up to 25 cm from the irradiated volume. For IMRT, internal scatter dominates only up to 10 cm from the irradiated volume, with collimator scatter dominating for the next 10 cm and head leakage beyond 20 cm.

The validation set is a more extensive set of measurements encompassing the range of clinical situations expected to be used at that facility. This will include depth dose and profile data at multiple SSDs, oblique angle data, the effect of inhomogeneity, and complex field shaping. Appendix 2 in this chapter shows suggested validation test measurements compiled from various sources including Appendix 3 of AAPM TG-53 and in International Atomic Energy Agency (IAEA) TRS-430, Commissioning and Quality Assurance of Computerized Planning Systems for Radiation Treatment of Cancer. Two other IAEA documents, TECDOC-1540 and TECDOC-1583, expand on TRS-430 and provide practical examples. The European Society of Therapeutic Radiation Oncology (ESTRO) and the Netherlands Commission on Radiation Dosimetry have also produced documents showing practical examples.

When performing these measurements many methods can be used. The most basic is the use of the appropriate detectors in a water phantom. Again, the details are described in Chapter 2 . The water phantom can acquire 1D and point dose data. Similarly, solid phantoms can be used in combination with point dosimeters. The point dosimeter is in actuality a (very small) volume dosimeter, so the user must ensure that the dosimeter is appropriate for the gradients and field sizes used.

By using a combination of water equivalent and non-water slabs, inhomogeneity effects can be analyzed. Because of the changes in beam spectra and scattering properties within or near non-water materials, corrections may need to be made to the dosimeter readings. For film embedded in non-water material the measured dose is the dose to the film, not dose to the inhomogeneity. The comparison can be made by using the dose to water for the calculation or applying the appropriate correction to the measurement using stopping power ratios. Ion chamber measurements will need similar corrections.

Film and detector arrays can be used to measure 2D data. Film has the advantage of much greater spatial resolution but can be problematic when trying to measure larger fields. Most radiotherapy departments no longer have film processors, so radiochromic film is used. The scanners used to analyze radiochromic films suffer from a lack of uniformity of response over larger areas, which limits the accuracy of the results. Detector arrays have a good uniformity of response over larger areas but suffer from poor spatial resolution. The detectors are situated 5 to 10 mm apart depending on the exact model used. There can also be directional response differences. Again these detectors can be used to evaluate inhomogeneity effects but are limited to transmission data since their design does not allow them to measure inside or proximal to an inhomogeneity.

Tissue Heterogeneity in the Context of TPS Dose Calculations

Some systems will use a CT number to density correction curve created using physical density while others use electron density. Not all dose calculation algorithms perform equally in and around tissue inhomogeneities, and the user should be aware of the algorithm uncertainties and limitations. AAPM TG-65, Tissue Inhomogeneity Corrections for Megavoltage Photon Beams, presents a detailed evaluation of inhomogeneity corrections. It describes the performance of different algorithms in various conditions and present recommendations. A summary of the recommendations is as follows:

• 1.

  Heterogeneity corrections should be applied to dose calculations and prescriptions. Even if there are inaccuracies in the algorithm, it will be closer to reality than if no correction is applied.

• 2.

  For head and neck, larynx, and lung treatments, even simple algorithms will calculate adequately in areas away from bone and air interfaces. Near these interfaces convolution and Monte Carlo algorithms will perform better. For high electron density materials, refer to AAPM TG-63, Dosimetric Considerations for Patients with Hip Prostheses Undergoing Pelvic Irradiation, for details.

• 3.

  For lung treatments energies of 12 MV or less are recommended.

• 4.

  For breast treatments, the doses near the chest wall/lung interface are more accurately calculated with convolution or Monte Carlo methods.

• 5.

  For gastrointestinal treatments, barium contrast areas should be contoured and the density should be overridden.

• 6.

  For prostate and pelvis treatments, the main concern is with high electron density hip prostheses. Again, see AAPM TG-63.

The calculation of dose to non-water structures is performed by using the CT numbers of the planning scan to account for differences in density. Typically a TPS will use a stored table to convert CT number to density. This table is established by scanning phantom materials of known density and correlating with the measured CT numbers. This CT–density relationship depends on the kVp of the scan and can be different from scanner to scanner even for the same kVp. The stability of this relationship must be verified over time and when the CT tube is replaced.

How would the tolerance levels be established for this QA measure? One way to approach this is to look at the typical clinical applications. Most tissue is near water equivalent, so to establish a clinically relevant range you could change the density of a water phantom and note when the calculated dose at 20 cm depth changes by more than 1%. For lower densities such as lung, a typical thickness is about 15 cm. Tolerance levels can be calculated by placing a calculation point distal to the lung volume and again changing the density until the dose changes by 1%. High density regions would not be expected to exceed 4 cm in thickness. Using this methodology yields ranges of +/−20 in CT number for low density to water and +/−100 for high density values. By comparison, AAPM TG-66, Quality Assurance for Computed-Tomography Simulators and the Computed-Tomography Simulation Process, recommends a generic +/−5 over the entire range of CT numbers.

Another issue with CT number to density conversions is the number of bits used to store the CT data, i.e., grayscale values. Using a traditional 12 bit storage will truncate the data at a CT number of approximately 3000, while 16 bit storage will allow much higher values. This will provide more accurate dosimetry and improved visualization.

Comparison of Calculated Doses to Benchmark Data

There are many sources of benchmark data in the peer-reviewed literature. These detail either a specific parameter such as output factors, a compilation of data, or test plans. Table 5.1 summarizes some of the literature. In addition, AAPM TG-23, Radiation Treatment Planning Dosimetry, has dosimetry data for characterization and validation, but it is somewhat old and may not have adequate data for some modern systems. It also does not have MLC data. IAEA TECDOC-1540 also has a set of benchmark data. Both data sets are available for download on the AAPM website, http://www.aapm.org/pubs/reports/ . Another older source of data is BJR Supplement No. 25, which contains depth dose data for various beam qualities. The data contain averages of many beams from different vendors, so care should be used when comparing to these data and a greater level of disagreement should be expected (2% to 5%). The Imaging and Radiation Oncology Core in Houston (IROC-H), formerly known as the Radiological Physics Center (RPC), has documentation on standard data for various linear accelerator models for both photon and electron beams. Their website also has many useful references ( http://rpc.mdanderson.org/RPC/home.htm ). AAPM TG-67, Benchmark Datasets for Photon Beams, produced a report describing the contents of a photon beam benchmark data set. Based on this report an Small Business Innovation Research (SBIR) grant was awarded to measure the data sets. Sun Nuclear Corpo­ration was awarded this grant and maintains the data. The vendor may also supply benchmark data. This should be used for comparison only and is not a substitute for local measurements. The use of vendor data as clinical data should only be done after validating the accuracy to within 1% to 2%.

Brachytherapy sources benchmark data are available for vendor-specific sources. The general calculation methodology is described in AAPM TG-43, Dosimetry of Interstitial Brachytherapy Sources, and Its Supplements.

Comparison of Calculated Doses From One System to Another System That Has Already Been Validated

This is a relatively simple process with several caveats. First, the limitations of the previous system must be known. For instance, differences would be expected when comparing a convolution model to a pencil beam model, and the comparison would not be valid for inhomogeneity calculations. In that case a comparison could be made for simple water calculations but measurements would have to be made for the more complex situations. Second, the capabilities of the two systems may not match. For example, the newer system may have volumetric arc therapy (VMAT) while the older does not. In the end, this type of comparison is of limited use for dosimetric commissioning. However, it is a valuable process to review with the clinical team any differences among the systems so that the results can be correlated with past experience. As mentioned previously, differences in inhomogeneity calculations should be understood as well as how penumbra differences may affect field size selection.

One of the most common examples of this is the verification of TPS MU calculation. AAPM TG-114, Verification of Monitor Unit Calculations for Non-IMRT Clinical Radiotherapy, describes this in detail. Typically the treatment plan data are transferred to the independent system using the Digital Imaging and Communications in Medicine (DICOM) protocol ( http://dicom.nema.org/ ). Depending on the system, the plan file, image files, and structure sets may be transferred. Some systems only calculate the MU based on an effective depth to the reference point and the treatment field fluence pattern. Other systems will perform a complete recalculation including dose-volume histograms (DVHs) for comparison. It is typical for second check systems to use the same input beam data as the commissioned TPS. This will speed up the implementation process; however, a rigorous QA of the second check system still needs to be performed.