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

DosimeTron, an agentic AI system powered by GPT-5.2, automates personalized Monte Carlo radiation dosimetry calculations for PET/CT medical imaging. Validated on 597 studies across 378 patients, the system achieved 99.6% correlation with reference dosimetry calculations while processing each case in approximately 32 minutes with zero execution failures.

DosimeTron represents a significant advancement in medical AI automation by successfully deploying agentic systems for complex, high-stakes clinical workflows. The system orchestrates multiple specialized tools through a reasoning engine that interprets natural language instructions, reducing the friction between clinical intent and computational execution. This approach demonstrates that autonomous AI agents can reliably handle intricate multi-step processes requiring coordination across DICOM processing, image segmentation, Monte Carlo simulation, and statistical analysis—domains traditionally requiring specialized expertise and manual oversight.

The healthcare industry has struggled to operationalize AI due to regulatory requirements, clinical validation demands, and the need for explainable decision-making. DosimeTron's achievement of zero hallucinations across diverse prompt configurations and perfect pipeline execution suggests that agentic AI architectures with proper tool sandboxing and monitoring can meet these stringent requirements. The 32-minute processing time makes the system clinically viable for routine use, positioning automated dosimetry as a practical alternative to manual calculations that consume specialist time.

This development carries implications beyond medical imaging. The successful implementation of agentic AI in a regulated, safety-critical domain validates the architectural patterns that combine language models with specialized tools and observability infrastructure. Healthcare institutions represent early adopters with budgets to implement such systems, potentially creating a precedent for deploying similar agentic workflows in finance, manufacturing, and other sectors requiring high accuracy and auditability. The technical feasibility demonstrated here suggests that the next phase involves scaling deployment, addressing regulatory certification pathways, and establishing best practices for monitoring agentic systems in clinical environments.

• →Agentic AI achieved 99.6% correlation with reference dosimetry across 114 test cases with zero execution failures or hallucinations.
• →The system autonomously coordinates DICOM processing, image segmentation, Monte Carlo simulation, and reporting through natural language instructions.
• →Clinical-scale processing times of 32 minutes per study demonstrate practical viability for routine medical use.
• →Zero-failure execution across diverse prompt configurations indicates robust agentic AI architecture suitable for regulated healthcare environments.
• →Successful validation in safety-critical medical dosimetry establishes proof-of-concept for agentic AI deployment in other high-stakes domains.