Click here to

Session: Multi-Disciplinary General ePoster Viewing [Return to Session]

Machine Learning Models to Predict the Delivered Positions of Elekta Multileaf Collimator Leaves for Volumetric Modulated Arc Therapy 

S Sivabhaskar1*, N Papanikolaou1, N Kirby1, R Li1, A Roy2, (1) University of Texas Health Science Center, San Antonio, TX, (2) The University of Texas at San Antonio, San Antonio, TX

Presentations

PO-GePV-M-35 (Sunday, 7/25/2021)   [Eastern Time (GMT-4)]

Purpose: To evaluate the performance of random forest (RF) regression, support-vector regression (SVR), and extreme gradient boosting (XGBoost) regression models in predicting the delivered positions of individual multileaf collimator (MLC) leaves.

Methods: In this study, 43 MLC log files containing 2380 control points for volumetric modulated arc therapy (VMAT) treatment plans delivered using an Elekta linear accelerator were obtained. A total of 8 planned parameters, namely gantry angle, collimator angle, X1 jaw position, X2 jaw position, leaf gap, leaf position, leaf velocity, and leaf acceleration were extracted. The RF, SVR, and XGBoost regression models were trained using 70% of the log files. The delivered leaf positions were used as the target variable for training, and a 6-fold cross validation method was used to validate and tune the model’s hyperparameters. The tuned models were tested on the remaining 30% of the MLC log files. The performance of the models was evaluated using mean absolute error (MAE), root mean square error (RMSE), R² value, and fitted line plots showing the relationship between delivered and predicted leaf positions.

Results: The RF model achieved a MAE of 0.427 mm, RMSE of 0.434 mm, and R² of 0.957. The SVR model achieved a MAE of 0.798 mm, RMSE of 0.809 mm, and R² of 0.849. The XGBoost model achieved a MAE of 0.214 mm, RMSE of 0.223 mm, and R² of 0.989.

Conclusion: Based on these results, the XGBoost model performed better than the RF and SVR models in predicting the delivered positions of the individual leaves for VMAT plans. Although the RF model had higher prediction errors, it still outperforms the SVR model. This study indicates that XGBoost has the potential to accurately predict the delivered MLC positions for VMAT treatment plans.

ePosters

    Keywords

    Not Applicable / None Entered.

    Taxonomy

    Not Applicable / None Entered.

    Contact Email

    Share: