LAUSR

This work proposes to use Artificial Neural Networks (ANN) for the regression of dosimetric quantities employed in mammography. The data were generated by Monte Carlo simulations using a modified and validated version of PENELOPE (v. 2014) + penEasy (v. 2015) code. A breast model of homogeneous mixture of adipose and glandular tissue was adopted. The ANN were constructed with Keras and scikit-learn libraries for Mean Glandular Dose (MGD) and Air Kerma (Kair) regressions, respectively. In total, seven parameters were considered, including the incident photon energies (from 8.25 to 48.75 keV), the breast geometry, breast glandularity and Kair acquisition geometry. Two ensembles of 5 ANN networks each were formed to calculate MGD and Kair. The Normalized Glandular Dose coefficients (DgN) are calculated by the ratio of the ensembles outputs for MGD and Air Kerma. Polyenergetic DgN values were calculated weighting monoenergetic values by the spectra bin probabilities. The results indicated a very good ANN prediction performance when compared to the validation data, with median errors on the order of the average simulation uncertainties (0.2%). Moreover, the predicted DgN values compared with works previously published were in good agreement, with mean(maximum) differences up to 2.2(9.3)%. Therefore, it was showed that ANN could be a complementary or alternative technique to tables, parametric equations and polynomial fits to estimate DgN values obtained via MC simulations.