SimSET and SimPET

Importance Sampling

SimPET configuration file offers the option to activate 3 importance sampling techniques that are avaliable in SimSET. These are the following:

• stratification: "true"
• forced_detection: "false"
• forced_non_absortion: "false"

Importance sampling is a way of accelerating convergence of the estimates of SimSET's histogrammed output. The three techniques are described bellow, but a more detailed description is provided in SimSET documentation (Importance Sampling)

Stratification

When stratification is used, the PHG chooses decay locations and initial photon directions that are likely to be detected with a higher probability than they would normally be sampled (figure 3). This is the most reliable importance sampling technique, but it requires 2 simulations to run (the first simulation is performed to calculate the probability of detection for each direction).

To activate stratification in SimPET, you should:

1. Set stratification: "true" in your config file
2. Set s_photons (sampling photons) to a value different than zero in your params file (this value will be used to perform the first simulation).

Forced detection

When forced detection is used, copies of photons are forced to scatter toward the tomograph somewhere within the volume visible to from the tomograph, and copies of photons within this volume are forced to pass through the attenuation to the tomograph (figure 4)

Forced non-absortion

With forced non-absorption, when photons interact in the object they always scatter - they are not allowed to be photoelectrically absorbed. This can be activated but it has very little impact on PET simulations (most interactions are scatter anyway)

Forced interaction (to be implemented in SimPET)

Forced-interaction causes every photon impinging on the first layer of the detector to undergo at least one interaction. The photon weight is adjusted to avoid bias (see Haynor et al, 1991). We suggest that you test this option for a given detector system before using it for longer runs: in some complex detector systems, particularly in complex block detector systems, it can slow simulations significantly. It provides the greatest detection efficiency gains in systems where photons are likely to pass through the detectors without interacting (e.g., animal PET systems with 10mm crystals or thin-crystal SPECT systems); it slows simulations down when photon paths can intersect many different detector elements in a complex detector system.