A planning quality evaluation tool for prostate adaptive IMRT based on machine learning.

Area of Science:

• Radiation Oncology
• Medical Physics
• Machine Learning Applications

Background:

• Adaptive intensity-modulated radiation therapy (IMRT) for prostate cancer requires robust plan quality evaluation.
• Ensuring optimal sparing of organs-at-risk (OARs) like the bladder and rectum is crucial in adaptive IMRT.
• Quantitative tools are needed to assess and improve the quality of adaptive radiation therapy (ART) plans.

Purpose of the Study:

• To develop a machine learning-based quantitative evaluation tool for adaptive IMRT prostate plan quality.
• To generate reference dose volume histograms (DVHs) for OARs using prior plans.
• To compare these reference DVHs with adaptive plans derived from fluence map deformation.

Main Methods:

• Developed a model using support vector regression (SVR) trained on 198 high-quality prostate IMRT plans.
• Characterized patient anatomy using organ volumes and distance-to-target histograms (DTH).
• Applied principal component analysis (PCA) to DVHs and DTHs to identify key features for the SVR model.

Main Results:

• Patient anatomical information was effectively quantified using only two or three principal components.
• The developed model demonstrated approximately 80% accuracy in predicting DVHs.
• The tool proved effective in enhancing the quality assessment of adaptive radiation therapy (ART) planning.

Conclusions:

• A machine learning-based tool for evaluating adaptive IMRT plan quality has been successfully developed.
• The tool estimates OAR sparing and provides a valuable reference for ART evaluation.
• This quantitative approach aids in ensuring better plan quality for adaptive IMRT of the prostate.