Improving the performance of direct Monte-Carlo optimization for large tumor volumes

Abstract

Direct Monte Carlo Optimization (DMCO) is a powerful method for dose optimization with Monte Carlo precision and direct aperture optimization with simulated annealing. In a recent publication we presented quasi-IMAT, a step-and-shoot technique that simulates a rotational method by using a high number of beam directions and reducing the number of segments. QIMAT could improve the plan quality, particularly in case of concave target volumes. In this work, both techniques were combined to optimize an anal cancer case. Because of the limited memory of standard computers, a technique for reducing the inverse kernel file was investigated on the CarPet phantom. It could be shown that increasing the statistical uncertainty from 5% to 15%, followed by a final dose calculation with 5%, did not lead to a decrease of the plan quality. Therefore, this method was applied to an anal cancer case optimization with qIMAT. The IK-file of a 30-field plan was calculated with 15% uncertainty and then used for optimization with DMCO (applying only 1 segment per field). After optimization, a re-calculation with 2% uncertainty was performed. In this way, the DMCO optimization with qIMAT of complex cases such as anal cancer is possible. We used the degeneracy of the solution space (similar dose distribution with different objective function values) to find an optimal solution of a complex case. The comparison of the qIMAT plan with a conventional 7-field IMRT plan showed that qIMAT can spare considerably dose to organs at risk and healthy tissue with uncompromised target coverage.