Texture analysis on the fluence map of intensity modulated beams to evaluate the plan delivery accuracy for quality assurance of patient treatment plans

Abstract

Purpose: The goal of this study is to evaluate the plan delivery accuracy of volumetric modulated arc therapy (VMAT) using texture analysis on the fluence map of VMAT. Methods: Twenty prostate and twenty head-and-neck (H&N) VMAT plans from TrilogyTM with MillenniumTM multi-leaf collimators (MLCs) (Varian Medical Systems, Palo Alto, CA, USA) were selected. Fluence maps were generated for each VMAT plan (1) by integrating spatial MU distributions defined by MLC apertures at each control point (CP), (2) by doubling the values (MUs) of pixels representing MLC tips, and (3) by taking summations of segments in sequential groups of CPs. A total of 6 textural features, which were angular second moment (ASM), inverse difference moment (IDM), contrast, variance, correlation, and entropy, were calculated from each fluence map generated from 20 prostate and 20 H&N VMAT plans. For each textural feature, the particular displacement distances (d) were 1, 5, and 10. To investigate the plan deliverability of each VMAT plan, gamma passing rates of pre-treatment quality assurance (QA), differences in modulating parameters such as MLC position, gantry angle, and monitoring unit (MU) between VMAT plans and dynamic log files registered by the linac control system during delivery, and differences in dose-volumetric parameters between VMAT plans and reconstructed plans using dynamic log files were acquired. To test the performance of textural features as indicators of modulation degree of VMAT plans, Spearmans rank correlation coefficients (rs) for the plan deliverability were calculated. For comparison purposes, conventional modulation indices for VMAT, i.e., the modulation complexity score for VMAT (MCSv), the leaf travel modulation complexity score (LTMCS), and the modulation index supporting station parameter optimized radiation therapy (MISPORT) were calculated

their correlations were analyzed using the same method. For the same 20 prostate and 20 H&N patients from TrilogyTM with MillenniumTM MLCs, VMAT plans were generated using TrueBeamTM with high-definition MLC (Varian Medical Systems, Palo Alto, CA, USA). Gamma passing rates of pre-treatment QA, differences in modulating parameters, and differences in dose-volumetric parameters were acquired. To compare the performances between the two MLC systems, fluence maps and a total of 6 textural features were calculated at different displacement distances (d) of 1, 2, 3, 5, and 10. Then, their correlations were analyzed using the same method. Results: In this study for the single fluence maps calculated using integrating spatial MU distributions defined by MLC apertures at each CP, there was no particular textural feature that consistently exhibited the best correlation for every type of plan deliverability. Taken together, the results for contrast (d = 1) and variance (d = 1) for TrilogyTM and variance (d = 1) for TrueBeamTM showed good correlations with every type of plan deliverability. These textural features showed consistently better performances than the conventional modulation indices, except for one case, namely, the modulating parameter differences. The rs values of contrast (d = 1) for TrilogyTM for the global gamma passing rates with criteria of 2%/2 mm, 2%/1 mm, and 1%/2 mm were 0.536, 0.473, and 0.718, respectively. The values of variance (d = 1) for TrilogyTM were 0.551, 0.481, and 0.688, respectively. The values of variance (d = 1) for TrueBeamTM were 0.662, 0.740, and 0.660, respectively. In the case of local gamma passing rates, the rs values of contrast (d = 1) for TrilogyTM were 0.547, 0.578, and 0.620, respectively. Those of variance (d = 1) for TrilogyTM were 0.519, 0.527, and 0.569, respectively. Those of variance (d = 1) for TrueBeamTM were 0.403, 0.616, and 0.443, respectively. Every rs value in these cases was statistically significant (p < 0.004). The values of rs of contrast (d = 1) and variance (d = 1) for TrilogyTM and variance (d = 1) for TrueBeamTM with MLC errors were -0.863, -0.828, and -0.822, respectively, with statistical significance (p < 0.001). These results most frequently showed rs values with statistical significance for the dose-volumetric differences. In this study, using fluence maps calculated by doubling the values (MUs) of pixels representing MLC tips, the rs values of contrast (d = 1) with edge-enhancement for TrilogyTM for global gamma passing rates of 2%/2 mm, 2%/1 mm, and 1%/2 mm were 0.546 (p < 0.001), 0.487 (p = 0.001), and 0.744 (p < 0.001), respectively. Those of contrast (d = 10) with edge-enhancement for TrueBeamTM were 0.705, 0.703, and 0.701, respectively (all with p < 0.001). The rs values of contrast (d = 1) with edge-enhancement for TrilogyTM for local gamma passing rates of 2%/2 mm, 2%/1 mm, and 1%/2 mm were 0.588, 0.644, and 0.640, respectively (all with p < 0.001). Those of contrast (d = 10) with edge-enhancement for TrueBeamTM were 0.459 (p < 0.001), 0.700 (p < 0.001), and 0.522 (p = 0.001), respectively. The rs values of contrast (d = 1) for TrilogyTM for MLC and gantry angle errors were -0.853 and 0.655, respectively (all with p < 0.001). Those of contrast (d = 10) for TrueBeamTM were -0.722 (p < 0.001) and 0.333 (p = 0.036), respectively. Contrast (d = 1) for TrilogyTM and contrast (d = 10) for TrueBeamTM showed statistically significant rs values in 11 and 4 dose-volumetric parameter differences, respectively, from a total of 35 cases. These contrasts showed generally better correlations with plan delivery accuracy than either (as previously suggested) textural features with non-edge-enhanced fluence maps or conventional modulation indices. In this study of fluence maps calculated using summations of segments at sequential groups of control points, the values of rs of contrast (d = 10) with 10 segments for TrilogyTM to both global and local gamma passing rates at 1%/2 mm were 0.692 (p < 0.001) and 0.798 (p < 0.001), respectively. The results showed rs values of -0.895 (p < 0.001) and 0.727 (p < 0.001) for multi-leaf collimator positional errors and gantry angle errors during delivery, respectively. The number of statistically significant rs values (p < 0.05) for the changes in dose-volumetric parameters during delivery was 14 from a total of 35 tested parameters. The values of rs of contrast (d = 10) with 20 segments for TrueBeamTM for both global and local gamma passing rates of 1%/2 mm were 0.839 (p < 0.001) and 0.738 (p < 0.001), respectively. The results showed rs values of -0.881 (p < 0.001) and 0.569 (p < 0.001) for multi-leaf collimator positional errors and gantry angle errors during delivery, respectively. The number of statistically significant rs values (p < 0.05) for the changes in dose-volumetric parameters during delivery was 8 among a total of 35 tested parameters. Conclusions: Contrast (d = 10) with 10 segments calculated from segmental fluence maps of VMAT plans for TrilogyTM and contrast (d = 10) with 20 segments calculated from segmental fluence maps of VMAT plans for TrueBeamTM showed good correlations with the plan deliverability, indicating that they could be used as indicators for assessing the delivery accuracy of VMAT plans.