|Abstract or Summary
- Monte Carlo (MC) algorithms are widely accepted as the most accurate method to calculate dose in a patient geometry. For this work the EGSnrc MC code was used as a benchmark for the identification of dose calculation errors produced by the widely implemented analytical anisotropic algorithm (AAA). By correlating the location and magnitude of these errors with the physical conditions under which AAA is known to fail, a set of error prediction methods was developed which can help to identify clinical plans that are at high risk for AAA dose calculation errors. Once these plans are identified, they can be recalculated with a more accurate algorithm. First, in order to calculate clinical treatment plans with MC, a treatment plan calculation framework (MCTPCF) was developed and validated. The underlying beam model used in the MCTPCF was thoroughly benchmarked against a standard open field data set. Radiochromic film measurements were then used to validate the geometry of the employed MC multileaf collimator (MLC) model. Mechanical functionality of the MCTPCF was verified by calculating several highly modulated clinical treatment plans and comparing them with AAA calculations. Next, three novel error prediction algorithms were developed and validated to a limited extent. The first, designated the field size index (FSI), identifies regions in the treatment plan space where many small fields or blocks overlap, leading to a build-up of beam modeling and volume averaging errors. The second, designated the heterogeneous scatter index (HSI), identifies regions within the electron density distribution where the AAA rectilinear kernel scaling approximation is stressed. The third, designated the low-density index (LDI), identifies regions of very low electron density where AAA is known to overestimate dose. An open field beam model for the 6MV Varian Clinac has been fully parameterized and is able to calculate dose to within 1.3% and 1.0 mm DTA (σ[mean] = 0.3%). The MCTPCF has been shown to accurately calculate highly modulated, multiple field treatments. FSI calculations show excellent agreement with MC/AAA deviations in highly modulated MLC fields in water, and to a lesser extent in patient geometry RapidArc treatments. The LDI accurately predicts AAA overdosing for simple geometries, however for the lung case investigated other sources of error made identifying any correlation a challenge. The theoretical structure of the HSI has been developed, however its implementation is still underway. An accurate MC based treatment plan calculation tool has been developed and validated. Three novel error prediction algorithms have been developed, two of which have been validated for homogenous geometries. In particular, the FSI shows promise as both a direct predictor of AAA error, and also as a general treatment plan complexity index. With sufficient benchmarking, these methods may be developed into a clinical tool that can identify treatment plans that are at high risk for AAA dose calculation errors.