A GPU-accelerated brachytherapy planning framework for template-based applications: Evaluated on three-dimensional...

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Imaging

Background:

• 3D-printed templates enhance High-Dose-Rate (HDR) brachytherapy for personalized cancer treatment.
• Current HDR brachytherapy planning faces limitations in optimization speed and needle design.

Purpose of the Study:

• To develop and validate a GPU-accelerated framework for efficient HDR brachytherapy planning.
• To improve plan quality, specifically target coverage and organ-at-risk (OAR) sparing, in template-based workflows.

Main Methods:

• A PyTorch-based framework with a GPU-accelerated dose engine and adaptive inverse optimizer was developed.
• Validation involved comparison with a commercial Treatment Planning System (TPS) using single-source tests and patient data.
• Retrospective re-optimization was performed on 13 cervical cancer patients treated with 3D-printed templates.

Main Results:

• The GPU dose engine demonstrated high accuracy, with dose differences below 1% compared to TPS in patient plans.
• The framework significantly improved High-Risk Clinical Target Volume (HR-CTV) conformity (+12.5%) and D100 values.
• Bladder OAR doses were reduced by ~2-3% (p<0.05), while rectum doses remained unchanged. Optimization time was reduced to 12-34 seconds per plan.

Conclusions:

• The GPU-accelerated framework offers accurate and rapid HDR brachytherapy planning.
• It enhances HR-CTV coverage and bladder sparing, crucial for cervical cancer treatment.
• The framework is suitable for integration into 3D-printed template workflows, supporting efficient, personalized care.