hybridMANTIS: a CPU-GPU Monte Carlo method for modeling indirect x-ray detectors with columnar scintillators

Area of Science:

• Medical Imaging
• Computational Physics
• Scientific Simulation

Background:

• Monte Carlo simulations for medical imaging require extensive computation time.
• Simulating indirect x-ray detectors necessitates high statistical accuracy, leading to long processing durations.

Purpose of the Study:

• To develop a novel hybrid approach for Monte Carlo simulations to accelerate the modeling of indirect x-ray detectors.
• To enhance the efficiency of computational modeling by maximizing CPU and GPU utilization.

Main Methods:

• Implemented a hybrid approach combining CPU and GPU computations for Monte Carlo simulations.
• Developed a GPU version of the fastDETECT2 optical transport model and a CPU version.
• Modified PENELOPE software for direct output of interaction data, integrated with fastDETECT2.
• Utilized a load balancer to dynamically allocate tasks to CPU and GPU cores.

Main Results:

• Achieved a speed-up factor of 627 compared to the original MANTIS code.
• Demonstrated a speed-up factor of 35 compared to a CPU-only version of the modified code.
• Successfully shifted the computational bottleneck from optical transport to x-ray transport.
• Enabled efficient simulation of large area detectors due to reduced memory requirements.

Conclusions:

• The hybrid Monte Carlo simulation approach significantly reduces computational time for indirect x-ray detector modeling.
• This method enhances the flexibility and efficiency of simulating complex detector geometries.
• The optimized simulation process allows for the study of large area detectors with greater ease.