Agentic AI supports personalized radiation dosimetry in PET/CT

Monday, November 30 | 9:30 a.m.-10:30 a.m. | M3-SSPH05-3 | Room E351

In this presentation, researchers will describe an agentic AI system that fully automates patient-specific Monte Carlo internal dosimetry on PET/CT exams.

Presenter Eleftherios Tzanis, PhD, of the University of Crete, and colleagues built the system by integrating a large language model with 23 software tools that handle tasks such as DICOM processing, organ segmentation, Monte Carlo simulation, and structured reporting. Users interact with it through natural language, with no programming expertise required.

The group tested the system on a public dataset of 597 PSMA PET/CT studies from 378 patients, using a range of single- and multi-turn prompts. They then compared its results with those of the established OpenDose3D software in 114 studies across 22 organs.

The researchers found that all workflows ran without errors. The system’s results closely matched the reference dosimetry, with a median correlation of 0.997 and a median difference of 2.5% across organs. Processing took about 32 minutes per study.

“This approach supports routine implementation of personalized radiation dosimetry in diagnostic and theranostic PET/CT by reducing workflow complexity, eliminating the need for programming expertise, and enabling computation of organ-level dosimetric quantities across diverse kinetic models,” the authors wrote.

Check out this session to get all of the details.

Page 1 of 2
Next Page