Using machine learning to predict gamma passing rate in volumetric-modulated arc therapy treatment plans

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Biology

Background:

• Volumetric-modulated arc therapy (VMAT) is a complex radiation therapy technique.
• Accurate prediction of treatment plan quality, such as gamma passing rate (GPR), is crucial for patient safety.
• Existing methods for VMAT plan evaluation can be time-consuming.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for predicting the gamma passing rate (GPR) in VMAT plans.
• To identify key plan complexity metrics that influence GPR prediction accuracy.
• To assess the performance of Random Forest Regression (RFR) and Support Vector Regression (SVR) models in predicting VMAT plan quality.

Main Methods:

• 118 clinical VMAT plans were analyzed using RayStation treatment planning system.
• Plan complexity metrics including plan-averaged beam area (PA), plan-averaged beam irregularity (PI), and total monitor units (MU) were computed.
• RFR and SVR models were trained on a 70% training dataset and tested on a 30% test dataset to predict GPR (3%/2mm, 10% threshold).

Main Results:

• Both RFR and SVR models demonstrated comparable performance in predicting GPR.
• The models achieved a root mean square error (RMSE) of approximately 1.4 for GPR prediction.
• Plan-averaged beam area (PA), plan-averaged beam irregularity (PI), and total monitor units (MU) were identified as the most influential metrics for GPR prediction.

Conclusions:

• Linear SVR and RFR models are effective and comparable for predicting GPR in VMAT plans.
• The identified complexity metrics (PA, PI, MU) can guide VMAT plan evaluation.
• This approach can potentially reduce uncertainties in radiation therapy planning and delivery.