Adaptive method for multicriteria optimization of intensity-modulated proton therapy

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Optimization

Background:

• Intensity-modulated proton therapy (IMPT) requires complex treatment planning to manage multiple competing objectives.
• Existing multicriteria optimization (MCO) methods face limitations in handling numerous objectives and approximating Pareto surfaces.
• Efficient generation of optimal treatment plans is crucial for maximizing therapeutic benefit and minimizing toxicity in IMPT.

Purpose of the Study:

• To develop and evaluate an adaptive multicriteria optimization (MCO) method for intensity-modulated proton therapy (IMPT) leveraging GPU acceleration.
• To address limitations of previous MCO approaches, specifically Pareto approximation and the number of objectives handled.
• To generate a dense and well-distributed Pareto surface for a large number of clinical objectives efficiently.

Main Methods:

• An adaptive MCO algorithm based on the augmented weighted Chebychev metric (AWCM) with a differential evolution (DE) adaptive weighting scheme was developed.
• The algorithm was implemented on a GPU cluster to enable parallel searches for rapid Pareto surface mapping.
• The method was tested on two clinical IMPT cases with 10 and 18 objectives, comparing MCO-generated plans against clinically optimized plans.

Main Results:

• The adaptive MCO algorithm efficiently generated Pareto points for cases with 10 and 18 objectives within 2-3 hours.
• The MCO approach produced a dense, well-distributed Pareto surface, minimizing the Pareto sampling metric.
• MCO-generated plans demonstrated significant improvements in DVH objectives, including substantial reductions in organ-at-risk (OAR) doses (e.g., 48-72% reduction in 50% dose).
• The AWCM method showed a clear dosimetric advantage over the weighted sum (WS) method.

Conclusions:

• An adaptive MCO algorithm accelerated by GPUs provides rapid and efficient mapping of the multicriteria Pareto surface for IMPT.
• The developed method generates clinically deliverable plans with improved dosimetric outcomes compared to conventional optimization.
• This GPU-accelerated adaptive MCO approach offers a promising advancement for complex IMPT treatment planning.