Reinforcement Learning in Medical Diagnosis: An Overview

Abstract

The paper provides the readers with the knowledge of how reinforcement learning (RL) applications can be applied in medical diagnosis and healthcare. RL is a powerful and mathematically sound semi-supervised approach that can be used to solve a plethora of medical problems. The paper opens up by providing the basic introduction to RL key components and digging deep into how basic reinforcement learning MDP framework can be applied to a medical problem. Since the treatment of a disease is affected by the individual’s past history and current course of action, these factors can be dynamically modeled by using sequential decision-making Markovian framework of RL. In addition to these, the paper also provides the insights into various open issues in RL and discusses the potential of RL in creation of dynamic treatment regimes (DTR) for automated medical diagnosis. RL has been efficiently used to develop DTR for the diagnosis of chronic diseases like cancer, HIV, diabetes, anemia and critical care. It is found that RL can be efficiently applied to a medical problem when the basic components of RL environment (like states, actions and rewards) are defined precisely.