OPTIMIZING INTENSITY-MODULATED RADIATION THERAPY ANGLE SELECTION USING GENETIC ALGORITHMS

摘要

Developing treatment plans for intensity-modulated radiation therapy (IMRT) involves first selecting beam angles and then optimizing the delivery of radiation from those angles. Optimizing the delivery of radiation for a set of beam angles is a time-intensive process and finding the beam angles from which to best deliver the radiation is an NP-hard problem. We use machine learning to construct models to quickly estimate the quality of optimized plans for new angle sets. These models are used as objective/fitness functions in a genetic algorithm (GA) that searches the space of feasible beam angle sets. Our GA quickly finds solutions for new patients based on the models trained on previously evaluated treatment plans for other patients. The solutions found are more than 2% better than the previously best-known solutions for patients.