Modeling complex particles phase space with GAN for Monte Carlo SPECT simulations: a proof of concept.

Area of Science:

• Medical Imaging
• Computational Physics
• Artificial Intelligence

Background:

• Monte Carlo (MC) simulations are crucial for emission tomography but computationally intensive.
• Accurate modeling of particle transport within patients is a major bottleneck in MC simulations.

Purpose of the Study:

• To develop a computationally efficient method for simulating particle transport in emission tomography.
• To accelerate the simulation of imaging systems, aiding in their design and optimization.

Main Methods:

• A generative adversarial network (GAN) was trained to model the distribution of particles exiting a patient.
• The GAN-generated particle distribution was integrated with a neural network for detector response modeling (ARF-nn).
• The combined approach was evaluated for single photon emission computed tomography (SPECT) simulations.

Main Results:

• The proposed GAN-based method significantly reduced simulation computation time compared to conventional MC methods.
• A complete rotating SPECT acquisition was simulated more efficiently.
• The method demonstrated feasibility for simulating various imaging systems and parameters.

Conclusions:

• Generative adversarial networks offer a powerful tool for accelerating emission tomography simulations.
• This approach enhances the efficiency of imaging system design and parameter exploration.
• The method shows promise for broader applications in medical physics and imaging research.