Efficient full Monte Carlo modelling and multi-energy generative model development of an advanced X-ray device

Area of Science:

• Medical Physics
• Computational Imaging
• Radiological Sciences

Background:

• Monte Carlo (MC) simulations for X-ray imaging require significant data storage for phase space (PhS) files, especially for medical applications.
• Generative Adversarial Networks (GANs) have been introduced to address the data storage challenges associated with traditional PhS files.

Purpose of the Study:

• To compare the efficacy of phase space (PhS) files, traditional Generative Adversarial Networks (GANs), and conditional GANs as photon sources in MC simulations against experimental measurements.
• To evaluate the performance of a conditional GAN in modeling multiple X-ray energies within a single network.

Main Methods:

• Modeled an X-ray imaging system (ImagingRing) using GATE-RTion v1.0 for MC simulations.
• Implemented traditional and conditional GAN models for PhS generation.
• Measured half-value layers (HVLs), focal spot, and Heel effect for validation.

Main Results:

• MC simulations using both GAN approaches showed agreement with measured HVLs, focal spot, and Heel effect.
• The conditional GAN, with specific regularization, performed comparably to the traditional GAN.
• GANs demonstrated significant advantages over PhS files in data storage and computational overhead.

Conclusions:

• Conditional GANs offer a more efficient and flexible approach for generating X-ray sources in MC simulations compared to traditional PhS files.
• The developed conditional GAN enables energy interpolation, decoupling network training from specific X-ray energy requirements.