Prediction of dosimetric accuracy for VMAT plans using plan complexity parameters via machine learning.

Area of Science:

• Medical Physics
• Radiation Oncology
• Machine Learning

Background:

• Volumetric Modulated Arc Therapy (VMAT) is a crucial radiotherapy technique.
• Ensuring dosimetric accuracy in VMAT plans is vital for effective cancer treatment.
• Predicting plan accuracy can streamline quality assurance processes.

Purpose of the Study:

• To predict the dosimetric accuracies of VMAT plans using machine learning models.
• To evaluate the effectiveness of plan complexity parameters in predicting VMAT plan accuracy.
• To compare the performance of regression tree analysis (RTA), multiple regression analysis (MRA), and neural networks (NNs) for this prediction task.

Main Methods:

• Utilized a dataset of 600 clinical VMAT plans.
• Included 28 predictor variables such as complexity parameters, machine type, and photon beam energy.
• Employed RTA, MRA, and NNs to predict passing rates for 5% dose difference (DD5%) and 3%/3 mm gamma index (γ3%/3 mm) using dosimetric measurements from a helical diode array (ArcCHECK).

Main Results:

• Mean passing rates for the entire dataset were 92.3% ± 9.1% for DD5% and 96.8% ± 3.1% for γ3%/3 mm.
• For the evaluation dataset (100 cases), prediction errors were minimal across all models.
• Neural networks (NNs) showed the lowest mean prediction errors: -0.2% ± 2.7% for DD5% and -0.2% ± 2.1% for γ3%/3 mm.

Conclusions:

• Neural networks exhibited slightly better prediction accuracy compared to RTA and MRA.
• The findings suggest that machine learning can effectively predict VMAT plan dosimetric accuracy.
• This approach holds promise for improving the efficiency of patient-specific quality assurance in radiotherapy.