Machine learning-based treatment couch parameter prediction in support of surface guided radiation therapy

Area of Science:

• Medical Physics
• Radiation Oncology
• Machine Learning Applications

Background:

• Surface-guided radiation therapy (SGRT) relies on precise patient positioning.
• Accurate treatment couch parameters are crucial for SGRT protocols.
• Manual verification of couch parameters can be time-consuming.

Purpose of the Study:

• To develop an independent, machine learning-based system for automatic prediction of treatment couch parameters.
• To support SGRT-based patient positioning protocols.
• To provide a quality assurance tool for patient positioning accuracy.

Main Methods:

• Utilized setup data from 183 patients across four groups based on setup devices.
• Calculated the difference between predicted and acquired treatment couch values.
• Developed a machine learning model for parameter prediction.

Main Results:

• Treatment couch parameters were predicted with high precision.
• A significant difference in variances was observed between Lung and Brain patient groups (p < 0.01).
• Prediction outliers were attributed to initial patient setup inconsistencies, not model inaccuracy.

Conclusions:

• Machine learning-based couch parameter prediction is accurate and serves as an effective starting point for SGRT.
• Verification of patient setup is necessary for large deviations (>1.5 cm) to optimize SGRT.
• The developed system enhances efficiency and accuracy in radiation therapy patient positioning.