Scout Monte Carlo: Preliminary Results of a Variance Reduction Technique for Photon Scattering Dose Calculation

While AI-based solutions have demonstrated significant benefits in enhancing the efficiency of Monte Carlo Simulations (MCS), several challenges persist. One notable issue is the requirement for extensive MCS to achieve acceptable statistical uncertainty, particularly in regions far from the source. This study presents 'Scout Monte Carlo' (Scout-MC), a variance reduction technique (VRT) designed to improve AI model training and facilitate rapid simulations for medical applications, such as out-of-field dosimetry. Scout-MC builds upon forced detection approaches and extends their capabilities to estimate deposited energy specifically at the voxel of interest, regardless of its location within the phantom. The method focuses on a sampling approach to calculate the probability of a particle reaching a specific voxel. Using these probabilities and the mass energy absorption coefficient, it approximates both fluence and deposited energy. For evaluation, a voxelized multimaterial phantom was targeted with a photon beam. The results were compared with standard MCS using GATE, revealing an average relative error of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3.4 \pm 32 \%$</tex>. Notably, Scout-MC demonstrated significantly lower and more uniform uncertainty (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1.9 \pm 0.91 \%$</tex>) compared to standard MCS (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$6 \pm 12 \%$</tex>). Scout-MC achieved these results in just 70 seconds, whereas GATE required 39 hours. Scout-MC can rapidly estimate the dose in any area of the image with low uncertainty and acceptable accuracy. Future work will involve extending the evaluation to more complex scenarios to further enhance accuracy and applicability.