Machine Learning-Based Prediction of Elekta MLC Motion with Dosimetric Validation for Virtual Patient-Specific QA.

Area of Science:

• Medical Physics
• Radiation Oncology
• Machine Learning in Healthcare

Background:

• Accurate multi-leaf collimator (MLC) motion prediction is crucial for advanced radiation therapy techniques like intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT).
• Traditional patient-specific quality assurance (QA) methods are resource-intensive and susceptible to measurement uncertainties.
• Developing predictive models can streamline QA processes and enhance dose delivery accuracy.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting delivered MLC positions on the Elekta Versa HD system.
• To assess the deterministic nature of the Elekta servo-mechanism using kinematic parameters from DICOM-RT plans.
• To evaluate the impact of ML-based motion prediction on dosimetric accuracy through physical QA measurements.

Main Methods:

• A dataset of 200 patient plans was created, pairing planned MLC kinematic parameters (positions, velocities, accelerations) with delivered values from trajectory logs.
• Four regression models, including linear regression (LR), were trained to predict delivered MLC positions.
• Dosimetric validation was performed using ArcCHECK measurements on 17 clinical plans, comparing gamma passing rates of original and LR-corrected plans.

Main Results:

• Linear regression (LR) achieved the highest prediction accuracy with a mean absolute error (MAE) of 0.145 mm, confirming a linear relationship between planned and delivered MLC trajectories.
• LR-corrected plans showed statistically significant improvements in gamma passing rates (mean increase of 2.24% at 1%/1 mm criterion, p < 0.001).
• The LR model effectively captured systematic mechanical signatures, such as inertial effects.

Conclusions:

• A computationally efficient linear regression model can accurately predict Elekta MLC performance, offering a robust method for ML-based virtual QA.
• This ML-driven approach reduces reliance on physical QA resources, making it valuable for time-sensitive workflows like adaptive radiotherapy (ART).
• Accurate MLC motion prediction enhances precision in dose delivery and optimizes QA efficiency in modern radiation oncology.