RadField3D: a data generator and data format for deep learning in radiation-protection dosimetry for medical...

Area of Science:

• Medical Physics
• Computational Physics
• Radiological Sciences

Background:

• Accurate radiation field data is crucial for dosimetry.
• Current simulation methods can be computationally intensive.
• Deep learning offers potential for faster radiation simulation.

Purpose of the Study:

• To develop an open-source Geant4-based Monte Carlo simulation application for generating 3D radiation field datasets.
• To introduce a machine-interpretable data format and Python API for integrating radiation data into deep learning research.
• To facilitate research into alternative radiation simulation methods using artificial intelligence.

Main Methods:

• Development of RadField3D, a Geant4-based Monte Carlo simulation application.
• Creation of a fast, machine-interpretable data format (RadFiled3D) with a Python API.
• Validation of simulation data against measured and simulated datasets.

Main Results:

• Successful generation of 3D radiation field datasets using the RadField3D application.
• Availability of a novel, efficient data format and API for neural network integration.
• Open-source release of all source codes and validation data for reproducibility.

Conclusions:

• RadField3D provides a valuable tool for generating radiation field data for dosimetry.
• The accompanying data format and API enable efficient use of this data in deep learning research.
• The open-source nature of the project promotes further research in AI-driven radiation simulation.