Abstract
PURPOSE: This study compares the dosimetric accuracy of deep-learning-based MR synthetic CT (sCT) in brain radiotherapy between the Analytical Anisotropic Algorithm (AAA) and AcurosXB (AXB). Additionally, it proposes a novel metric to predict the dosimetric accuracy of sCT for individual post-surgical brain cases. MATERIALS AND METHODS: A retrospective analysis was conducted on 20 post-surgical brain tumor patients treated with Volumetric Modulated Arc Therapy (VMAT). sCT and planning CT images were obtained for each patient. Treatment plans were optimized on sCT and recalculated on planning CT using both AAA and AXB. Dosimetric parameters and 3D global gamma analysis between sCT and planning CT were recorded. The bone volume ratio, a novel metric, was calculated for each patient and tested its correlation with gamma passing rates. RESULTS: For AAA, the mean differences in D(mean) and D(max) of PTV between sCT and planning CT were 0.2% and -0.2%, respectively, with no significant difference in PTV (p > 0.05). For AXB, mean differences in D(mean) and D(max) of PTV were 0.3% and 0.2%, respectively, with significant differences in D(mean) (p = 0.016). Mean gamma passing rates for AXB were generally lower than AAA, with the most significant drop being 9.3% using 1%/1 mm analyzed in PTV. The bone volume ratio showed significant correlation with gamma passing rates. CONCLUSIONS: Compared to AAA, AXB reveals larger dosimetric differences between sCT and planning CT in brain photon radiotherapy. For future dosimetric evaluation of sCT, it is recommended to employ AXB or Monte Carlo algorithms to achieve a more accurate assessment of sCT performance. The bone volume ratio can be used as an indicator to predict the suitability of sCT on a case-by-case basis.