Analytical modeling and feasibility study of a multi-GPU cloud-based server (MGCS) framework for non-voxel-based dose...

Area of Science:

• Radiotherapy
• Computational Physics
• Cloud Computing

Background:

• Radiotherapy requires computationally intensive dose calculations.
• Online adaptive therapies demand faster treatment planning.
• Existing hardware limitations hinder advanced computational methods.

Purpose of the Study:

• To present a multi-GPU cloud-based server (MGCS) framework for radiotherapy dose calculations.
• To explore remote computing for parallelization and acceleration of radiotherapy tasks.
• To assess the feasibility of MGCS for online adaptive therapies.

Main Methods:

• Developed an analytical model for MGCS performance prediction and workload distribution.
• Conducted numerical studies using 14 GPUs across 4 servers (1 Gbps network).
• Optimized inter-process communication to minimize data transfer and facilitate resource distribution.

Main Results:

• Analytical predictions closely matched experimental observations (1-5% difference).
• MGCS performance scaled with the number of GPUs for compute-intensive tasks.
• Accurate dose computations were reproduced with negligible differences.

Conclusions:

• A cloud-based computation engine is a feasible solution for rapid dose calculations in advanced radiotherapy.
• The MGCS framework offers significant acceleration for computationally intensive tasks, outperforming local machines.
• This approach democratizes access to advanced radiotherapy technology and computational methods for clinics.