A deep learning-based approach for statistical robustness evaluation in proton therapy treatment planning: a...

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence

Background:

• Robustness evaluation in particle radiotherapy is crucial due to treatment uncertainties.
• Current methods using limited uncertainty scenarios lack statistical consistency.
• Accurate dose prediction is essential for reliable treatment planning.

Purpose of the Study:

• To develop an artificial intelligence (AI)-based method for predicting percentile dose distributions in particle radiotherapy.
• To enable robust evaluation of planning objectives at specific confidence levels.
• To overcome limitations of traditional robustness evaluation methods.

Main Methods:

• A deep learning (DL) model was trained to predict 5th and 95th percentile dose distributions.
• Predictions were based on nominal dose distributions and planning CT scans.
• Ground truth percentile doses were estimated using 600 dose recalculations per patient.

Main Results:

• DL predictions showed excellent agreement with ground truth percentile doses (mean error < 0.15 Gy).
• DL achieved high gamma passing rates (GPR) (>93.9% at 1 mm/1%), significantly outperforming worst-case scenario (WCS) evaluations (<54%).
• Dose-volume histogram analysis confirmed DL's superior accuracy and precision over WCS.

Conclusions:

• The proposed AI method provides fast and accurate percentile dose predictions for radiotherapy robustness evaluation.
• This approach enhances statistical interpretation and confidence in treatment planning.
• The AI-driven method has the potential to significantly improve radiotherapy robustness assessment.