An Ensemble Approach to Knowledge-Based Intensity-Modulated Radiation Therapy Planning.

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Biology

Background:

• Knowledge-based planning (KBP) uses prior treatment plans to predict optimal dose volume histograms (DVHs) for new patients.
• Regression models in KBP translate patient anatomical features into organ-at-risk DVH predictions.
• Existing regression methods show variable performance depending on the clinical context.

Purpose of the Study:

• To develop a more robust knowledge-based planning (KBP) model for radiation therapy.
• To improve the accuracy and reliability of dose volume histogram (DVH) predictions.
• To create an ensemble method combining multiple regression techniques for enhanced KBP.

Main Methods:

• An ensemble approach was developed, integrating stepwise, lasso, elastic net, and ridge regression models.
• Individual model prediction metadata were generated using in-training-set leave-one-out cross-validation.
• Constrained optimization determined model weights, and metadata filtered training set outliers.

Main Results:

• The proposed ensemble method demonstrated improved robustness compared to individual regression models.
• The ensemble approach was more resilient to small training set sizes and data inaccuracies.
• Evaluations on prostate and head and neck intensity-modulated radiation therapy (IMRT) cases confirmed enhanced performance.

Conclusions:

• The ensemble KBP method offers superior robustness, making it more suitable for clinical application.
• This approach enhances the reliability of DVH predictions in radiation therapy planning.
• The findings suggest a significant advancement in the clinical utility of knowledge-based planning.