A comparison of three optimization algorithms for intensity modulated radiation therapy.

Area of Science:

• Medical Physics
• Radiotherapy
• Computational Optimization

Background:

• Intensity modulated radiation therapy (IMRT) and intensity modulated proton therapy (IMPT) rely on optimization algorithms for treatment field modulation.
• Current in-house tools often use quasi-Newton algorithms, which, despite good results, have limitations.

Purpose of the Study:

• To implement and evaluate an improved optimization algorithm based on the limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) routine.
• To compare the performance of the L-BFGS algorithm against standard quasi-Newton and conjugate gradient methods in IMRT and IMPT planning.

Main Methods:

• The study describes an improved optimization algorithm utilizing the L-BFGS routine.
• Treatment plans for both photon (IMRT) and proton (IMPT) therapies were optimized using the new algorithm, the standard quasi-Newton algorithm, and the conjugate gradient algorithm.
• Performance was evaluated based on speed and objective function value improvement.

Main Results:

• The L-BFGS-based algorithm was, on average, six times faster than the standard quasi-Newton algorithm in reaching the same objective function value.
• The improved algorithm consistently achieved lower objective function values, with an average improvement of 37%.
• The conjugate gradient algorithm showed an average speedup factor of two and a 30% improvement in the objective function value, ranking between the other two algorithms.

Conclusions:

• The implemented L-BFGS optimization algorithm offers significant acceleration and improved plan quality in intensity modulated radiotherapy.
• The benefits are particularly notable in proton therapy (IMPT) plans.
• This enhanced optimization approach can lead to more efficient and effective radiation treatment planning.