Automatic Planning of Whole Breast Radiation Therapy Using Machine Learning Models.

Area of Science:

• Radiation Oncology
• Medical Physics
• Computational Biology

Background:

• Whole breast radiation therapy (WBRT) is a standard treatment for breast cancer.
• Current WBRT planning is time-consuming and relies on manual adjustments.
• Optimizing treatment planning can improve efficiency and patient outcomes.

Purpose of the Study:

• To develop an automated treatment planning system for WBRT.
• To enable near-real-time planning using intensity-modulated tangential fields.
• To optimize energy selection and dose distribution for WBRT.

Main Methods:

• Developed an automatic energy prediction model using PCA of DRR histograms.
• Utilized a random forest model for initial fluence map generation.
• Implemented automatic anchor point selection and centrality correction for dose balance.

Main Results:

• The auto-planning system achieved comparable plan quality to manual WBRT plans.
• Energy selection accuracy was high, matching clinical choices in 19/20 cases.
• Planning time was drastically reduced to under 20 seconds from hours.

Conclusions:

• An automatic WBRT planning system was successfully developed.
• The system provides clinically comparable plan quality with optimal energy selection.
• Significant reduction in planning time allows for near-real-time WBRT delivery.