Photon effective dose prediction using XGBoost: A machine learning approach for radiological protection

The International Commission on Radiological Protection (ICRP) developed dose conversion coefficients (DCC) using effective dose calculated for anthropomorphic phantoms based on human organs physical characteristics. Advances in computational technology and Monte Carlo codes have refined these models. Machine Learning (ML), increasingly applied in radiation physics, shows promise for enhancing dose prediction in radiation protection and personalized dosimetry by handling large DCC datasets. This study collected photon DCC data from various phantoms, representing diverse demographics, and prepared it through cleaning and segmentation. The eXtreme Gradient Boosting (XGBoost) model, optimized with Bayesian methods, was used to predict organ and effective doses. Results showed high accuracy for energies above 30 keV, with KERMA DCC yielding lower Mean Squared Error and fluence DCC exhibiting higher R2 values. However, predictions at lower energies were less accurate for both sets. This work highlights ML's potential to revolutionize personalized dosimetry by providing a fast alternative to Monte Carlo simulations. Future research should refine predictions for lower energy ranges and incorporate additional features to further enhance model accuracy.