Evaluation of a Quantitative Dose Comparison Tool for IMRT and Conformal Therapy

Introduction: We evaluated a dose comparison tool that can be used to evaluate measured and calculated dose distributions(1). The tool can be used to evaluate dose calculation algorithms by comparison against measurements, or calculations against calculations. The tool is based on a generalization of the dose-difference and distance-to-agreement tools recently described by Harms et al.(2) and is a measurement of the disagreement between a reference distribution and an evaluated distribution where the reference distribution can be as sparse as a single point or as large as a volumetric dataset, and the evaluated distribution is at least a one-dimensional dataset. The tool is computed independently for each reference point and is evaluated in a generalized space consisting of the distance and dose difference between the reference point and the evaluated distribution points. The distance and dose axes are normalized to a distance-to-agreement and dose-difference test criteria, respectively, so the axes are unitless. The minimum distance in this normalized space is labeled by the symbol γ. The comparison passes if γ≤1 and fails if γ>1. The original description of this tool appeared in Low et al.(1) which contained a simple example based on a onedimensional dose distribution intended to mimic a megavoltage beam open field profile. An unadjusted profile was used as the reference distribution, while the evaluated distribution was the same profile altered to model a positional offset, a dosimetric offset, and a combination of the two. The original study did not address the influence of the pixel size, criteria selection, or measurement noise on the results. This presentation will detail our work on evaluating the influence of these criteria on the selectivity of the tool. The test experiments are all conducted in two dimensions and use a model of an open 10 x 10 cm megavoltage field (Figure 1) with 2.0 x 2.0 mm pixels. The field is divided into four quadrants, and the reference distribution remains unaltered. The evaluated distribution, however, is adjusted as a function of the quadrant (Figure 2). Quadrant 1: Reference and evaluated distributions equal Quadrant 2: Evaluated distribution shifted in position orthogonal to the field edge Quadrant 3: Evaluated distribution is modified by an additive dose error as a function of the off-axis distance parallel to the field edge, providing an evaluation of γ for variations of dose magnitude, rather than position. Quadrant 4: Combination of the dose and position shifts is applied. Figure 3 shows the dose difference between the reference and evaluated distributions. Figure 4 shows the value of γ with 3% and 3 mm criteria. Because two-dimensional distributions are difficult to quantitatively evaluate, a cumulative histogram is prepared as a function of γ. Pixel Size: In regions of high dose gradient, the value of γ can be incorrectly determined because of the rapid change in dose as a function of position. For example, in a one-dimensional case if two distributions are exactly the same, but are offset by a half a pixel with width x ∆ , the calculation of γ will take place in the neighboring pixel