Fast 3D whole-body occupational dose estimation in interventional radiology using physics-informed deep learning.

Area of Science:

• Medical Physics
• Radiology
• Artificial Intelligence

Background:

• Occupational radiation exposure in interventional radiology is complex and not well-measured by current methods.
• Traditional dosimetry struggles with the spatially varying nature of scattered radiation.

Purpose of the Study:

• To develop a fast, physics-informed deep learning framework for predicting 3D physician dose distributions.
• To overcome limitations of conventional dosimetry in interventional radiology.

Main Methods:

• Utilized GPU-accelerated Monte Carlo simulations to generate 3D dose maps.
• Trained residual and transformer-based 3D U-Net architectures on simulated data.
• Evaluated model performance using error metrics, gamma analysis, and personal dose equivalents.

Main Results:

• The residual 3D U-Net model achieved high accuracy with mean absolute errors < 0.06 nGy and >90% gamma passing rates.
• Predicted personal dose equivalents closely matched Monte Carlo references.
• Demonstrated anatomically resolved dose estimation for various tissues.

Conclusions:

• The proposed framework enables rapid, accurate, and detailed occupational dose estimation.
• This technology can enhance real-time radiation awareness and guidance systems.
• Potential for significant improvements in operator radiation protection in interventional procedures.