GMMC-IR: a GPU-accelerated fast dose calculation algorithm for interventional radiology.

Area of Science:

• Medical Physics
• Computational Imaging
• Radiological Sciences

Background:

• Interventional radiology (IR) procedures involve prolonged fluoroscopy, posing radiation risks to patients and operators.
• Accurate dosimetry is crucial for managing radiation exposure in complex IR scenarios.
• Traditional Monte Carlo (MC) simulations are computationally intensive, limiting their clinical application.

Purpose of the Study:

• To develop and validate GMMC-IR (GPU mesh Monte Carlo for interventional radiology), a GPU-accelerated MC code for rapid dosimetry in IR.
• To maintain high physical fidelity in dose assessment for IR procedures.

Main Methods:

• Developed GMMC-IR using NVIDIA CUDA kernels and OptiX™ for ray tracing, optimizing for GPU performance.
• Employed an event-based approach to address GPU thread divergence and incorporated memory pooling and dynamic scheduling.
• Validated GMMC-IR against Geant4 simulations using a water phantom, the MRCP_AF phantom, and realistic IR scenarios.

Main Results:

• GMMC-IR demonstrated dose calculation differences within 3% for depth and lateral profiles in a water phantom compared to Geant4.
• Organ dose differences in complex phantoms and IR scenarios were within 5%.
• Achieved significant speedup factors ranging from 90 to 855 compared to the Geant4 baseline.

Conclusions:

• GMMC-IR offers a viable solution for rapid dose distribution assessment in interventional radiology.
• The developed tool can aid in improving radiation safety for both patients and healthcare professionals during IR procedures.