Deep learning-based synthetic dose-weighted LET map generation for intensity modulated proton therapy.

Area of Science:

• Medical Physics
• Radiation Oncology
• Artificial Intelligence

Background:

• Proton therapy's Bragg peak enhances tumor targeting but current planning overlooks variable relative biological effectiveness (RBE).
• The fixed RBE of 1.1 in clinical practice does not account for proton RBE variability, impacting treatment precision.
• Accurate dose-average linear energy transfer (LETd) calculation is crucial for optimizing proton therapy but is computationally intensive.

Purpose of the Study:

• To develop and validate a deep learning framework for predicting LETd distribution maps from dose distribution maps.
• To simplify and accelerate the generation of LETd maps for clinical proton therapy planning.
• To improve the accuracy of proton therapy by better incorporating variable RBE effects.

Main Methods:

• A deep learning framework utilizing a CycleGAN model was developed to predict LETd maps from dose maps.
• The model's performance was evaluated using metrics such as Mean Absolute Error (MAE), Peak Signal-to-Noise Ratio (PSNR), and Normalized Cross Correlation (NCC).
• Validation was performed within the clinical target volume, bladder, and rectum to assess clinical relevance.

Main Results:

• The proposed CycleGAN model achieved high accuracy in predicting LETd maps, with an overall MAE of 0.096 ± 0.019 keV/μm.
• The model demonstrated excellent performance with a PSNR of 24.203 ± 2.683 dB and NCC of 0.997 ± 0.002.
• Specific MAE values in critical structures were 0.193 ± 0.103 keV/μm (clinical target volume), 0.277 ± 0.112 keV/μm (bladder), and 0.211 ± 0.086 keV/μm (rectum).

Conclusions:

• The deep learning framework successfully generates synthetic LETd maps from dose maps, demonstrating feasibility for clinical application.
• This approach has the potential to significantly improve proton therapy planning by providing rapid and accurate LETd information.
• The findings support the integration of AI-driven tools to enhance the precision and effectiveness of proton therapy treatments.