Article: A Framework for Prediction of Personalized Pediatric Nuclear Medical Dosimetry Based on Machine Learning & Monte Carlo Techniques

This work introduces a methodology for using Artificial Intelligence (AI) techniques to develop an internal dosimetry prediction toolkit for nuclear medical pediatric applications. Based on unique anatomical features, Monte Carlo simulations were run as a ground truth to precisely predict absorbed doses per organ in pediatric patients. Using a population of computational pediatric models using GATE Monte Carlo simulations, the study produced a simulated dosimetry database for 28 pediatric phantoms and 5 radiopharmaceuticals. Using this dataset, machine learning regression models were built through the application of ensemble learning and hyperparameter optimization approaches. The resulting toolkit can predict specific absorbed dose rates (SADRs) in 30 organs for five different radiopharmaceuticals in pediatric patients with high accuracy (<2.7% uncertainty, >90% accuracy), providing fast results (<2 seconds). The method is applicable to other medical dosimetry applications in different patient populations.