Difference between matRad_calcParticleDoseMC and matRad_calcDoseDirectMC

[ZengYL99]

Hi,

I ran the matRad_example12_simpleParticleMonteCarlo.m in matRad-dev_varRBErobOpt using the MCsquare engine. The calculation was fast when utilizing resultGUI_MC = matRad_calcDoseDirectMC(ct,stf,pln,cst,resultGUI.w). However, the calculation significantly slowed down when using dij = matRad_calcParticleDoseMC(ct, stf, pln, cst).
I have checked the dij = matRad_calcParticleDoseMCsquare(ct,stf,pln,cst,calcDoseDirect), and found that the primary difference in setting calcDoseDirect to true or false before starting MCsquare is the modification of stfMCsquare(i).energyLayer(k).MU. Could this be the reason for the slower computation when using matRad_calcParticleDoseMC compared to matRad_calcDoseDirectMC?

Best regards,
Zeng.

[wahln]

Both functions are doing different things for different purposes.
matRad_calcDoseDirectMC does a forward dose calculation, i.e., takes a plan with optimized weights and runs a Monte Carlo simulation. It will directly return a dose cube within matRad result structure resultGUI. As the beamlet weights are known, significantly less histories are needed (as some beamlets will have very low weights) than in the second case:
matRad_calcParticleDoseMC will compute the dose influence matrix, returned as the dij struct, which can be used for optimization to determine the weights. As the weights are not known yet, each beamlet of the plan needs to be calculated to sufficient accuracy, requiring significantly more MC histories than the forward calculation.

[wahln]

One more comment: The calcDoseDirect functions as an internal switch to change the behavior of the dose accumulation (either in a dose cube, or in a dose influence matrix).

[ZengYL99]

Thank you for your reply. It is very helpful for me!
Additionally, I have another question regarding MCsquare. It seems that matRad supports parallel calculations for Monte Carlo simulation as it configured the following parameters:

pln.propMC.parallelRuns = true;
pln.propMC.numThreads = 8;

However, the parallel computing remains inactive after configuring these parameters. Do you have any insights or suggestions on this issue? Your assistance in this matter would be highly appreciated.

[wahln]

parallelRuns has no meaning for our MCsquare interface, it is only considered by TOPAS.
MCsquare is tuned for multi-core CPUs and handles highly-optimized multicore parallelization internally. If your CPU still is idling on some cores, increase the number of threads.
if you want to run separate simulations on multiple nodes / CPUs, this is not really supported within our MCsquare interface.

[ZengYL99]

I got it! Thank you very much! :)