Inverse optimization of objective function weights for treatment planning using clinical dose-volume histograms

Area of Science:

• Radiation Oncology
• Medical Physics
• Computational Biology

Background:

• Clinical treatment planning relies on optimizing radiation dose delivery to target tumors while sparing healthy tissues.
• Dose-volume histograms (DVHs) are critical tools summarizing dose distributions but extracting planning objectives from them is challenging.
• Existing inverse optimization (IO) methods may not fully capture clinical intent or can lead to complex treatment plans.

Purpose of the Study:

• To develop and validate a novel inverse optimization (IO) model for estimating objective function weights directly from clinical DVHs.
• To generate 'inverse plans' using these estimated weights and compare their quality and characteristics to original clinical plans.
• To assess the feasibility of integrating this IO approach into current clinical workflows for improved efficiency and plan quality.

Main Methods:

• Developed a novel IO model to derive objective function weights from clinical DVHs.
• Applied the model to 217 head and neck cancer treatment plans from Princess Margaret Cancer Centre.
• Compared DVH metrics, planning criteria satisfaction, and dose differences between original clinical plans and newly generated inverse plans.

Main Results:

• Inverse plans closely matched clinical DVHs, with median dose differences within 1.1 Gy.
• Clinical planning criteria satisfaction showed minimal differences (≤1.4%) between clinical and inverse plans.
• The novel IO approach yielded plans comparable or superior to existing methods, with lower fluence heterogeneity and fewer undesirable features like hotspots.

Conclusions:

• Clinical DVHs contain sufficient information to estimate objective function weights for generating high-quality, clinically relevant radiation therapy plans.
• The developed IO model offers a promising method for improving treatment planning efficiency and initializing the planning process.
• This approach can be integrated into knowledge-based planning and adaptive radiotherapy frameworks for automated plan generation.