SU-E-T-611: Utilizing Machine Learning Techniques for Beam Angle Selection in Radiation Treatment Planning

Area of Science:

• Medical Physics
• Radiation Oncology
• Machine Learning

Background:

• Intensity-modulated radiation therapy (IMRT) is a cornerstone of modern radiation oncology.
• Optimizing beam angles is crucial for maximizing treatment efficacy and minimizing dose to healthy tissues.
• Current methods for beam angle selection can be computationally intensive and may not always yield optimal results.

Purpose of the Study:

• To develop and evaluate a machine learning (ML)-driven approach for optimizing Intensity-modulated radiation therapy (IMRT) beam angle selection.
• To determine an optimal set of IMRT beam angles using ML techniques integrated with global optimization.
• To improve upon existing methods for beam angle selection in IMRT planning.

Main Methods:

• Utilized a dataset of pre-generated Intensity-modulated radiation therapy plans (e-plans) with 72 potential beam angles.
• Developed a beam set scoring function based on overdose/underdose criteria.
• Employed the Nested Partitions (NP) global optimization framework combined with ML for rapid, approximate scoring of beam angle sets.
• Extracted dose data from individual beams to train ML models for predicting dose components and overall scores.

Main Results:

• Achieved average improvements of 43%, 29%, and 11% in plan scores compared to default, best e-plans, and conventional NP methods, respectively.
• Demonstrated significant improvements in organ sparing, including 10% for the spinal cord, 12% for the brain stem, and 15% for the oral mucosa.
• The ML-enhanced NP approach identified superior beam angle sets more efficiently than traditional methods.

Conclusions:

• Machine learning tools offer an effective method for rapid, high-quality approximate scoring of beam angle sets in IMRT.
• Integrating ML with the NP global optimization framework leads to the selection of excellent beam angle sets.
• This ML-driven optimization has the potential to enhance IMRT treatment planning and patient outcomes.