Point cloud dosimetry framework for preclinical microbeam radiation therapy.

Area of Science:

• Medical Physics
• Radiotherapy Research
• Computational Dosimetry

Background:

• Microbeam Radiation Therapy (MRT) is an experimental radiotherapy technique requiring accurate dose calculation for clinical advancement.
• Monte Carlo (MC) simulations are the gold standard for dosimetry but are computationally intensive, hindering clinical use.

Purpose of the Study:

• To address the slow execution times of MC simulations for MRT dose prediction.
• To explore the efficacy of 3D point cloud regression models as a scalable alternative to voxel-based methods for MRT dosimetry.
• To evaluate the performance of the SphereFormer model in replicating MC dosimetry.

Main Methods:

• The SphereFormer model, a 3D point cloud-based network, was trained to emulate gold-standard Monte Carlo simulations.
• The model was validated on uniform voxel grids, assessing its accuracy in dose prediction.
• The scalability of the point cloud approach was demonstrated using out-of-field information.

Main Results:

• The SphereFormer model achieved high accuracy in MRT dose prediction, replicating MC dosimetry.
• The method demonstrated significant improvements in valley dose prediction accuracy (within 3% for 84.1% of voxels) compared to baseline methods (78.2%).
• The utilization of out-of-field information enhanced prediction accuracy, showcasing the model's scalability.

Conclusions:

• This study presents a proof-of-concept for using 3D point cloud methods in MRT dose prediction.
• This research marks the first application of point cloud-based methods to general dosimetry tasks.
• The findings suggest a promising, scalable, and accurate approach for future radiotherapy dose calculations.