A machine learning approach to the accurate prediction of multi-leaf collimator positional errors.
Joel N K Carlson1, Jong Min Park, So-Yeon Park
1Program in Biomedical Radiation Sciences, Department of Transdisciplinary Studies, Graduate School of Convergence Science and Technology, Seoul National University, Seoul 08826, Korea. Biomedical Research Institute, Seoul National University Hospital, Seoul 03080, Korea.
Physics in Medicine and Biology
|March 8, 2016
Summary
Machine learning accurately predicts multi-leaf collimator (MLC) discrepancies in radiotherapy, improving dose accuracy and quality assurance. Predicted positions enhance dose calculations, leading to more realistic treatment plans and better patient outcomes.
Area of Science:
- Medical Physics
- Radiotherapy Technology
- Machine Learning Applications
Background:
- Discrepancies between planned and delivered multi-leaf collimator (MLC) positions introduce errors in radiotherapy dose distributions.
- Accurate prediction of MLC motion is crucial for reliable treatment delivery and quality assurance (QA).
Purpose of the Study:
- To develop and validate machine learning models for predicting MLC positional discrepancies during radiotherapy delivery.
- To assess the impact of predicted MLC positions on QA procedures, dosimetry, and patient dose distributions.
Main Methods:
- Trained machine learning models using parameters from DICOM-RT plan files (e.g., leaf position, velocity) and DynaLog files from QA delivery.
- Calculated predictive leaf motion parameters, including direction relative to the MLC isocenter.
- Incorporated predicted MLC positions into treatment planning system (TPS) dose calculations and compared with measured dose distributions.
Main Results:
- The developed model accurately predicts MLC positions during delivery, with predicted positions being closer to delivered positions than planned positions for moving leaves.
- Incorporating predicted positions into dose calculations improved gamma passing rates in QA measurements (e.g., 4.17% increase for head and neck plans with 1%/2 mm criteria).
- Dose-volume histograms (DVHs) calculated using predicted positions showed closer agreement to delivered DVHs compared to planned DVHs, especially for peripheral organs at risk.
Conclusions:
- Machine learning models can accurately predict MLC motion discrepancies in radiotherapy.
- Using predicted MLC positions in dose calculations provides a more realistic representation of delivered dose, enhancing QA and treatment planning.
- This approach offers a more accurate view of the actual dose distribution received by the patient, aiding in treatment plan optimization.
