A GPU-accelerated framework for individualized estimation of organ doses in digital tomosynthesis

Area of Science:

• Medical Physics
• Radiological Sciences
• Computational Imaging

Background:

• Organ dose estimation in digital tomosynthesis (DT) is complex due to challenges in modeling specific collimations and source-detector trajectories.
• Existing methods often lack flexibility for various exam protocols and suffer from computational inefficiencies in Monte Carlo (MC) simulations.

Purpose of the Study:

• To develop and benchmark a GPU-accelerated MC simulation framework for accurate, individualized organ dose estimation in DT.
• To overcome limitations of existing tools by incorporating patient-specific computational phantoms and flexible protocol modeling.

Main Methods:

• A two-step workflow was developed: a MATLAB code to compute patient-specific fields of view (FOVs) and an MC-GPU tool to estimate organ doses.
• The framework was used to estimate doses for 28 radiosensitive organs in an adult male phantom using a commercial DT system.
• Benchmarking was performed against a reference dataset for a chest exam, quantifying relative errors and timing performance.

Main Results:

• The developed framework demonstrated close agreement with the reference dataset, with a mean absolute percent difference of 1.7% for most organs.
• Higher relative errors were observed for testes (-18.9%) and eye lens (-27.6%) due to their positioning outside the primary irradiation field.
• The simulation run time was significantly reduced to 916.3 seconds, indicating substantial performance improvement over nonparallelized MC tools.

Conclusions:

• A validated GPU-accelerated framework for patient-specific organ dose estimation in DT has been successfully developed.
• This tool facilitates radiation burden tracking for dose monitoring and supports the optimization of DT exam protocols.
• The framework provides essential capabilities for clinicians and researchers to enhance radiation safety in DT procedures.