Machine learning-based modeling of the anode heel effect in x-ray Beam Monte Carlo simulations

Area of Science:

• Medical Physics
• Computational Imaging
• Machine Learning

Background:

• The anode heel effect causes asymmetric x-ray beam intensity, impacting imaging system simulations.
• Accurate modeling of this effect is crucial for realistic Monte Carlo (MC) simulations in medical imaging.

Purpose of the Study:

• To develop a machine learning (ML) framework for modeling the anode heel effect in MC simulations.
• To enable accurate, energy-dependent beam modeling with minimal experimental data.

Main Methods:

• Trained multiple regression models, including gradient boosting regression (GBR), to predict spatial intensity variations.
• Used experimentally acquired weights from beam measurements and a fine-tuning protocol.
• Implemented GBR models in OpenGATE and GGEMS MC toolkits.

Main Results:

• GBR achieved the highest accuracy with prediction errors under 5% across energy levels.
• An optimized fine-tuning strategy reduced measurement effort by 65% (6 positions/energy level).
• ML-based models closely replicated clinical beam profiles, outperforming symmetric models.

Conclusions:

• Presents a robust ML method for incorporating the anode heel effect into MC simulations.
• Enhances simulation realism for dosimetry, image quality, and radiation protection applications.
• Facilitates accurate beam modeling with limited calibration data.