To improve computational efficiency, researchers from UT Southwestern Medical Center and LMU Munich have developed goCMC – a graphics processing unit (GPU)-oriented MC package for carbon ion therapy. The goCMC tool performs faster simulations in part due to its efficient approach to handling nuclear interaction simulation.

"Unlike a general purpose Monte Carlo package, for example Geant4, which uses complicated hadronic models, goCMC uses tabulated data to support nuclear interaction sampling," explained lead authors Nan Qin and Marco Pinto. "These tables were built according to Geant4 simulation results and are thus reliable, since Geant4 is regularly benchmarked against experiments. When goCMC is executed, these tables are read into memory of the computing device. Querying data from memory is much faster than performing complicated calculations based on hadronic models."

The new tool also uses a high degree of parallelization to speed simulations. To achieve this, the team developed goCMC using the OpenCL framework, which enables parallel computing on different devices, including GPUs and multi-core CPUs (Phys. Med. Biol. 62 3682).

Code validation

To validate goCMC, the researchers performed dose calculations at various beam energies and compared the results with Geant4-computed doses. They examined four homogeneous phantoms (water, lung, soft tissue and bone), an inhomogeneous phantom (a water phantom with a half-lung, half-bone slab) and a patient case. Overall, they observed good agreement between goCMC and Geant4 for dose distributions and range estimations.

In the water phantom, depth-dose curves for 100 and 250 MeV/u beams matched well, whereas small discrepancies existed for the 400 MeV/u beam in front of the peak region. Small discrepancies were also seen when comparing profiles beyond the Bragg peak, mainly due to the fact that goCMC ignores tertiary nuclear interaction products contributing to dose beyond the Bragg peak.

For the patient case, the researchers simulated a 250 MeV/u beam impinging laterally on a patient head CT scan. Here, in addition to Geant4 simulation without transport of neutral particles and secondary electrons, they also performed a full Geant4 simulation. Comparing dose profiles revealed good agreement of between goCMC and the two Geant4 simulations. This also confirmed that neglecting neutral particle and secondary electron transport is a valid approach.

To quantify discrepancies, the researchers computed voxel-by-voxel dose differences between goCMC and Geant4 (in regions receiving more than 10% of the maximum dose). The mean difference relative to the maximal dose was 1% or less for all except the low-energy lung case, which showed a difference of 1.6%.

From the depth-dose curves, they also determined differences in beam range (the distal position where dose dropped to 80% of the peak value) between goCMC and Geant4 for the homogeneous phantoms. Most range differences were well below 1 mm, except for a larger value of 1.36 mm for 400 MeV/u beam in lung. This extreme case corresponds to over 50 cm range and is not clinically relevant.

The researchers also performed 3D gamma tests on the goCMC-calculated doses, using the corresponding Geant4 data as reference. With a 1%/1mm criterion, the passing rate (within the 10% isodose line) was over 90% in all but two extreme cases. Using a less strict criterion of 2%/1 mm, all cases achieved a passing rate of over 96%.

Clinical potential

To test the efficiency and portability of goCMC, the team ran simulations on various GPUs and CPUs. The resulting dose distributions were consistent with each other. The authors note that goCMC could achieve a clinically acceptable dose calculation accuracy within 3–200 seconds on a single GPU, depending on the beam energy and GPU type.

The team is now developing a fully MC-based treatment planning system for intensity-modulated carbon ion therapy. "We have recently finished developing biological dose calculation and biological inverse treatment plan optimization," said project lead Xun Jia and Katia Parodi. "Next, we are going to implement beam models according to realistic beam data. Upon completion, we will perform comprehensive evaluations of the system." Meanwhile, they are also working on validation of positron emitting nuclei distributions and prompt gamma distributions calculated by goCMC, for use in range verification.

Related articles in PMB
Initial development of goCMC: a GPU-oriented fast cross-platform Monte Carlo engine for carbon ion therapy
Nan Qin et al Phys. Med. Biol. 62 3682
Recent developments and comprehensive evaluations of a GPU-based Monte Carlo package for proton therapy
Nan Qin et al Phys. Med. Biol. 61 7347
A GPU OpenCL based cross-platform Monte Carlo dose calculation engine (goMC)
Zhen Tian et al Phys. Med. Biol. 60 7419

Related stories

• Dose engine brings MC calculations to all
• Monte Carlo codes under close scrutiny
• Shrinking uncertainty in carbon ion therapy
• GPU accelerates proton dose calculations