Distributed and scalable optimization for robust proton treatment planning.

Area of Science:

• Medical Physics
• Computational Biology
• Radiotherapy

Background:

• Robust proton treatment planning is crucial for mitigating uncertainties.
• Current methods face computational challenges due to the increasing number of uncertainty scenarios.
• This increases the overall time required for treatment planning.

Purpose of the Study:

• To develop a fast and scalable distributed optimization platform for robust proton treatment planning.
• The platform parallelizes computations across multiple uncertainty scenarios.
• This aims to reduce the computational cost and planning time.

Main Methods:

• Modeled robust proton treatment planning as a weighted least-squares problem.
• Employed the alternating direction method of multipliers with Barzilai-Borwein step size (ADMM-BB).
• Reformulated the problem into parallelizable subproblems for each uncertainty scenario, evaluated on head-and-neck cancer patients.

Main Results:

• ADMM-BB generated robust plans with comparable or superior dosimetric quality to projected gradient descent (PGD).
• ADMM-BB achieved a 6-7 times faster average runtime compared to PGD.
• The speedup improved with an increasing number of uncertainty scenarios.

Conclusions:

• ADMM-BB is an effective distributed optimization method for accelerating robust proton treatment planning.
• It utilizes parallel processing platforms like multicore CPUs and GPUs.
• This leads to shorter planning times and improved plan quality through the inclusion of more uncertainty scenarios.