Reinforcement learning based control of tumor growth with chemotherapy

In this paper, optimal drug schedule for patients in progressive cancer phase who take the drug through infusion pump is obtained. An objective of control is reducing tumor cell numbers effectively while minimizing total amount of drug regimen. This is done because of the known serious side effects and major damages resulting from chemotherapy. Chemotherapy brings about weakness of the patient´s immune system which is one of the most dangerous side effects. The optimal control problem is to design an effective drug-schedule to reduce the size of the tumors in a time-optimal fashion. To achieve this goal, a reinforcement learning (RL), which is one of the best unsupervised machine learning algorithms, is proposed for control. Because RL has no need of environment model, i.e. it is model-free; it has absorbed interests during the recent year, especially in medical applications. Performance evaluation of the proposed algorithm has been performed by simulating on the mathematical model of tumor cells interacting with immune system. Simulation results show that a burst of treatment at the beginning is the best way to battle the tumor and constant decreasing the dosage of drug let the immune system to be reconstructed.