Voxel-based internal dose prediction using machine learning with organ-specific features and Monte Carlo simulations

Abstract

Estimation of internal dose is a critical task in nuclear medicine and radiation protection. New organ-specific features are included to construct a machine learning model capable of predicting the internal dose in the UF/NCI voxel phantoms, ranging from newborn to pediatric and adult of both genders. The dosimetry data is generated using the Monte Carlo simulation toolkit, Gate. Multiple source organs were utilized to train and validate the predictive models. Results demonstrate high accuracy, with less than 2% Root Squared Error in predicting the internal dose in most organs using the XGBoost machine learning model. This research can help nuclear medicine and radiation protection researchers and practitioners refine internal dose predictions based on anatomical and physiological characteristics of patients.