Annihilation photon GAN source model for PET Monte Carlo simulation

Area of Science:

• Medical Imaging
• Computational Physics

Background:

• Positron Emission Tomography (PET) imaging relies heavily on Monte Carlo simulations for accurate modeling.
• Current simulation methods are computationally intensive, limiting research and development.

Purpose of the Study:

• To develop a Generative Adversarial Network (GAN) model for efficient simulation of gamma emissions in PET imaging.
• To generate back-to-back gamma pairs with timing information for enhanced Monte Carlo simulations.

Main Methods:

• A conditional GAN was trained using low-statistic simulation data of gamma rays exiting an attenuation phantom.
• A novel parameterization was introduced to improve GAN training efficiency.
• The generated data was evaluated using an ideal PET reconstruction algorithm and NEMA/IEC phantoms.

Main Results:

• The GAN accurately reproduced proportions of 2-gammas, 1-gammas, and absorbed gammas within 1% of reference simulations.
• Image profiles and recovery coefficients showed less than 5% difference compared to reference data.
• The GAN demonstrated a tendency to slightly blur the 511 keV gamma energy peak.

Conclusions:

• The proposed GAN model effectively generates realistic gamma emission data for PET simulations.
• This approach significantly accelerates Monte Carlo simulations, offering speedups up to 400x.
• The trained GAN can serve as a fast, efficient source for PET imaging system simulations.