DosimeTron: Automating Personalized Monte Carlo Radiation Dosimetry in PET/CT with Agentic AI

A research team has developed and evaluated a novel artificial intelligence system named DosimeTron, which automates personalized Monte Carlo radiation dose calculations for patients undergoing PET/CT scans, as detailed in a study published on the arXiv preprint server under identifier arXiv:2604.06280v1. This agentic AI system was created to address the complex, time-consuming, and manual nature of patient-specific internal dosimetry, which is critical for optimizing radiation safety and treatment planning in nuclear medicine. The system represents a significant advancement in computational medical physics by employing a sophisticated agentic architecture. Its core is powered by GPT-5.2, which acts as a reasoning engine to orchestrate a suite of 23 specialized tools. This allows DosimeTron to autonomously execute the multi-step workflow required for Monte Carlo simulations, a gold-standard method for accurately modeling how radiation deposits energy in human tissues. The evaluation was conducted retrospectively using a substantial, publicly available dataset of 597 PSMA-PET/CT studies from 378 male patients. These studies utilized two common radiotracers, Fluorine-18 (369 studies) and Gallium-68 (228 studies), and were acquired across three different scanner models, testing the system's robustness and generalizability. The successful development of DosimeTron could herald a new era of efficiency and standardization in clinical dosimetry. By automating this intricate process, the technology promises to reduce human error, free up valuable expert time for clinical decision-making, and make highly personalized dose assessments more routinely accessible. This is particularly vital in fields like theranostics, where precise radiation dose estimates are directly linked to treatment efficacy and patient safety. The publication of this work on arXiv facilitates rapid dissemination to the scientific community, inviting further validation and collaboration to translate such AI-driven tools from research into practical clinical application.