Three-dimensional dose and LET_D prediction in proton therapy using artificial neural networks.

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Biology

Background:

• Proton therapy offers advantages but faces challenges in patient selection, plan consistency, and biological uncertainties related to linear energy transfer (LET).
• Knowledge-based planning (KBP) presents a potential solution to optimize proton therapy by leveraging data-driven approaches.

Purpose of the Study:

• To develop and evaluate the first three-dimensional (3D) dose and dose-weighted LET (LET _D ) prediction model for cranial proton radiation therapy using artificial neural networks.
• To assess the accuracy of the model in predicting dose and LET _D distributions compared to actual treatment plans.

Main Methods:

• Artificial neural networks were trained on 117 cranial proton therapy treatment plans, including dose and LET _D data.
• The dataset was divided into training, validation, and testing sets.
• Model performance was evaluated using dose- and LET _D -volume metrics and Dice similarity coefficients (DSC) for isodose lines.

Main Results:

• The developed model demonstrated high accuracy in predicting dose distributions, with dose-volume metrics showing minimal significant differences compared to planned doses.
• Dice similarity coefficients (DSC) for 50%, 75%, and 95% isodose lines were favorable (0.90, 0.93, and 0.88, respectively).
• The model also showed good agreement for LET _D -volume metrics, supporting its utility for biological dose considerations.

Conclusions:

• A novel 3D dose and LET _D -prediction model for cranial proton therapy has been successfully developed.
• The model's dose prediction accuracy is comparable to existing models for other treatment sites.
• The agreement in LET _D predictions supports the use of KBP for incorporating biological dose considerations in proton therapy, potentially enhancing treatment efficacy and patient outcomes.