IMRT QA using machine learning: A multi-institutional validation.

Area of Science:

• Medical Physics
• Radiation Oncology
• Machine Learning Applications

Background:

• Intensity-Modulated Radiation Therapy (IMRT) requires rigorous Quality Assurance (QA) to ensure treatment accuracy.
• Current IMRT QA methods can be time-consuming and resource-intensive.
• Virtual QA approaches offer a potential solution for efficient and accurate IMRT verification.

Purpose of the Study:

• To validate a machine learning (ML) based Virtual IMRT QA framework.
• To assess the framework's accuracy in predicting gamma passing rates.
• To evaluate performance across different measurement techniques and institutions.

Main Methods:

• A previously developed ML algorithm was tested on independent IMRT measurements from two institutions.
• Institution 1 used diode-array detectors; Institution 2 used portal dosimetry.
• A weighted Poisson regression with Lasso regularization predicted passing rates using 90 plan complexity metrics.

Main Results:

• The ML model predicted gamma passing rates within 3% accuracy at Institution 1.
• Accuracy was within 3.5% for 120/139 plans at Institution 2 using portal dosimetry.
• Key predictive features included irradiated area, jaw position, and MLC leaf gaps.

Conclusions:

• Virtual IMRT QA successfully predicts passing rates across different measurement techniques and institutions.
• This ML-driven approach has significant potential to enhance current IMRT QA processes.
• Accurate prediction of QA passing rates can streamline IMRT workflow and improve safety.