Fast and accurate ray-tracing simulation of multi-pinhole Molecular Breast Tomosynthesis

Objectives Imaging of radiolabelled molecule distributions is gaining popularity for detecting breast tumours. Recently, we proposed a novel design for dedicated molecular breast tomosynthesis (MBT) based on sliding multi-pinhole collimators [1]. Fast photon transport simulations play a key role in the design, optimization, and image reconstruction of such novel molecular imaging systems. Here we propose and validate Voxelized Ray Tracer (VRT) software for MBT.

Methods VRT uses a voxelized model of the collimator as its input, where each voxel is assigned an attenuation coefficient. With such a versatile voxelized collimator model, collimator geometries that can be simulated are not restricted to simple geometric shapes. For efficient raytracing, a multi-resolution approach is taken; initial raytracing is performed on coarse voxels followed by raytracing on a much finer voxel grid in locations with small structures. Additionally, VRT has the possibility to incorporate either continuous or pixelated gamma detectors and can include a Gaussian detector resolution model and variable depth of interaction in the gamma detector. While attenuation in collimator and gamma detector are taken into account, gamma photon scatter is neglected for sake of computational speed. To assess VRT’s accuracy for MBT, we compared a series of 99mTc point-spread functions (PSFs) with both VRT and the well-validated Monte-Carlo package GATE [2]. Additionally, as VRT does not model Compton scattered photons from the torso and especially the heart and liver, which might degrade MBT images, we determined their scatter fraction in the 99mTc photo peak. For this scatter simulation, a MCAT torso phantom with heart, liver, and breast compartments was used with relative uptake retrieved from literature on 99nTc-MIBI.

Results The PSFs generated by followed the PSFs of GATE very closely, with a mean relative error of only 0.5% and a maximum error of 4.8% relative to the PSF’s maximum. Yet, VRT needed less than 1/1000th of the computation time of the GATE simulation. Furthermore, we found that in the photo peak window (20% width centred at 140 keV) only about 1.5 % of the counts were due to scatter from the torso, which can be explained by the fact that photons from the torso require a significant change of direction and thus significant loss of energy to be able to pass through the pinholes in the MBT geometry.

Conclusions Dedicated VRT allows accurately calculating photon transport for MBT at only a fraction of the computational costs of a full GATE simulation. Combined with a tool that converts STEP-files to voxelized volumes, VRT can be used to quickly evaluate a large variety of new collimator-detector geometries.

• Download figure
• Open in new tab
• Download powerpoint