Optimizing Interstitial Photodynamic Therapy Planning With Reinforcement Learning-Based Diffuser Placement.

Area of Science:

• Oncology
• Medical Physics
• Computational Biology

Background:

• Interstitial photodynamic therapy (iPDT) is a promising minimally invasive cancer treatment.
• Improved photosensitizers enhance iPDT efficacy, but personalized treatment planning remains a challenge.
• Current iPDT lacks optimized planning for light source parameters considering patient anatomy and predicted response.

Purpose of the Study:

• Develop novel algorithms for personalized iPDT treatment planning.
• Optimize light source positions and powers to maximize tumor eradication and minimize damage to organs-at-risk.
• Incorporate accurate light propagation modeling using Monte Carlo simulations.

Main Methods:

• Utilized simulated-annealing as a baseline for source placement.
• Introduced novel source perturbation strategies to enhance treatment outcomes.
• Implemented a reinforcement learning-based method to optimize algorithm runtime and perturbation selection.
• Simulated algorithm performance on virtual brain tumors (glioblastoma multiforme) using 5-ALA PpIX photosensitizer.

Main Results:

• The developed algorithm achieved an average of 46% reduction in damage to organs-at-risk compared to manual placement.
• Optimized planning demonstrated superior outcomes in simulated glioblastoma cases.
• Reinforcement learning effectively reduced computational runtime without compromising plan quality.

Conclusions:

• A general and high-quality iPDT planning system can significantly improve treatment effectiveness.
• This advancement broadens the applicability of iPDT to diverse oncological indications.
• The developed planning system is poised to facilitate further clinical trials and adoption of iPDT.