Double Q-Learning for Intelligent Multi-Drug Scheduling in Cancer Chemotherapy Optimisation

Area of Science:

• Computational Biology
• Artificial Intelligence in Medicine
• Pharmacology

Background:

• Chemotherapy scheduling is complex, balancing tumor suppression with patient toxicity.
• Existing methods struggle to dynamically adapt to individual patient responses and drug resistance.

Purpose of the Study:

• To develop and evaluate a double Q-learning controller for optimizing daily dosing of a three-drug chemotherapy regimen.
• To minimize tumor burden while adhering to strict clinical toxicity constraints.

Main Methods:

• A pharmacokinetics-pharmacodynamics (PK/PD) tumor model with eight resistance states was used as a simulation environment.
• A double Q-learning algorithm was implemented to optimize daily drug dosages (cisplatin, docetaxel, irinotecan).
• Controller performance was evaluated against classical Q-learning and assessed for robustness under parameter variations and disturbances.

Main Results:

• Double Q-learning achieved near-complete tumor suppression (residual tumor fraction $~10^{-6}$) within the simulation.
• The controller effectively maintained systemic toxicity within predefined clinical constraints.
• Robustness analyses confirmed stable closed-loop behavior and resilience to physiological parameter variations (up to ±50%) and disturbances.

Conclusions:

• Double Q-learning offers a significant advancement over classical Q-learning for chemotherapy optimization.
• This AI-driven approach provides a proof-of-concept for adaptive chemotherapy, demonstrating potential for future clinical applications.
• Reinforcement learning frameworks show promise for personalized and effective cancer treatment strategies.