Room: AAPM ePoster Library
Purpose: To fast and accurately evaluate the impact of a patient’s respiratory motion on dose distributions of scanned proton beams.
Methods: In order to perform fast and accurate dose calculations for continuously changing patient geometries characterized by 4D-CT, a GPU-based Monte Carlo code has been integrated into an open source DICOM-RT Ion interface (http://github.com/topasmc/dicom-interface). This integration enabled not only to process dose calculations and accumulations on-the-fly as a function of breathing phases but also to support various DICOM inputs, such as 4D-CT, DICOM-RT Ion plan, and DICOM-REG for deformation vector fields (DVF). For example, the RTI interface calculates the time structure of a beam delivery sequence from a DICOM planand provides primary protons per each breathing phase for GPU-based dose calculations. Then, the simulated doses are accumulated using the DVF.
Results: The performance of the streamlined calculations was tested for a liver patient, e.g., 5 seconds breathing cycle with 10 phases in the 4DCT (0.5 seconds time-resolution), 10 DICOM-REG files for the DVF, and 40 seconds for total beam delivery. The test includes assessment of the execution time required in addition to the GPU simulation time during total beam delivery, e.g., data-transfer time between CPU (Xeon E5-2620) and GPU (Quadro K620) and dose-accumulations. For updating the patient geometries on the GPU and the dose accumulations, it took a total of 45±2 seconds. Primaries are generated and transferred to the GPU at 0.35±0.03M histories per second. Thus, a total of 180 seconds and 45 seconds were required for CPU operations to simulate 50M and 5M histories on the GPU, respectively.
Conclusion: By integrating a GPU-based MC system into a DICOM interface software, 4D-CT dose calculations became easy to use and efficient. This streamlined 4D-CT dose calculation will help assessment of motion sensitive proton radiotherapy for treatment planning and quality assurance purposes.