An AI dose-influence matrix engine for robust pencil beam scanning protons therapy

Area of Science:

• Medical Physics
• Radiation Oncology
• Artificial Intelligence in Medicine

Background:

• Proton pencil beam scanning (PBS) therapy necessitates rapid planning to address range uncertainties.
• Calculating the dose influence matrix (D_ij) for robust PBS plan optimization is computationally intensive and slow.

Purpose of the Study:

• To develop and integrate an AI-D_ij engine into a treatment planning system (TPS) to accelerate D_ij calculations in PBS therapy.

Main Methods:

• An AI-D_ij engine utilizing a transformer-based spot dose calculation model (SDM) was developed.
• The SDM processes CT volumes and energy levels to output spot dose distributions, trained on extensive data.
• Clinical plans were replanned using the AI-D_ij engine and compared against existing TPS, with dose evaluation via Monte Carlo (MC) methods.

Main Results:

• AI-D_ij engine optimization showed minimal differences in DVH metrics for clinical target volumes (CTV) and organs at risk (OARs) compared to Raystation.
• MC dose evaluation also demonstrated comparable results between the AI-D_ij engine and Raystation plans.
• Robust AI-D_ij calculations were achieved in under 2 seconds on an RTX3090 GPU.

Conclusions:

• The AI-D_ij engine presents a feasible solution for robust PBS plan optimization.
• The engine demonstrates significant improvements in planning speed without compromising plan quality.