Optimal Fractionation Scheduling for Radiotherapy Treatments with Reinforcement Learning, Tumor Growth Modeling and...

Area of Science:

• Medical Physics
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• Radiotherapy is a cornerstone of cancer treatment, involving fractionated radiation doses to balance tumor targeting and minimize healthy tissue damage.
• Current radiotherapy planning faces challenges due to inaccuracies in tumor growth models, necessitating adaptive and robust treatment strategies.

Purpose of the Study:

• To enhance radiotherapy treatment planning using reinforcement learning (RL) for improved adaptability and robustness.
• To investigate the efficacy of RL in minimizing healthy tissue damage while achieving tumor eradication across different cancer sites.

Main Methods:

• A 2D tumor growth simulation model was developed to train RL agents.
• Tabular RL techniques were employed to determine optimal radiotherapy treatment strategies.
• The Lyman Normal Tissue Complication Probability (NTCP) model was integrated to predict and assess treatment-related complications.

Main Results:

• The RL approach demonstrated significant reductions in healthy tissue damage: 10.7% for rectal, 49.1% for head and neck, and 37.5% for lung cancers compared to baseline treatments.
• RL-based strategies successfully achieved tumor eradication across all simulated sites.
• The study successfully analyzed treatment outcomes and complications in simulated rectal, head and neck, and lung cancer sites.

Conclusions:

• Reinforcement learning offers a powerful tool for optimizing radiotherapy treatment planning.
• RL-based radiotherapy significantly reduces healthy tissue damage and improves treatment efficacy compared to traditional methods.
• This AI-driven approach holds promise for developing more personalized and effective cancer therapies.