Reinforcement learning-based control of tumor growth under anti-angiogenic therapy

In recent decades, cancer has become one of the most fatal and destructive diseases which is threatening humans life. Accordingly, different types of cancer treatment are studied with the main aim to have the best treatment with minimum side effects. Anti-angiogenic is a molecular targeted therapy which can be coupled with chemotherapy and radiotherapy. Although this method does not eliminate the whole tumor, but it can keep the tumor size in a given state by preventing the formation of new blood vessels. In this paper, a novel model-free method based on reinforcement learning (RL) framework is used to design a closed-loop control of anti-angiogenic drug dosing administration.

A Q-learning algorithm is developed for the drug dosing closed-loop control. This controller is designed using two different values of the maximum drug dosage to reduce the tumor volume up to a desired value. The mathematical model of tumor growth under anti-angiogenic inhibitor is used to simulate a real patient.

The effectiveness of the proposed method is shown through in silico simulation and its robustness to patient parameters variation is demonstrated. It is demonstrated that the tumor reaches its minimal volume in 84 days with maximum drug inlet of 30 mg/kg/day. Also, it is shown that the designed controller is robust with respect to  ± 20% of tumor growth parameters changes.

The proposed closed-loop reinforcement learning-based controller for cancer treatment using anti-angiogenic inhibitor provides an effective and novel result such that with a clinically valid and safe dosage of drug, the volume reduces up to 1mm³ in a reasonable short period compared to the literature.