Automatic Disease Diagnosis System Using Deep Q-Network Reinforcement Learning

Disease diagnosis is a pivotal and indispensable step in the overall treatment process of diseases. The rapid advancement of machine learning and deep learning methodologies has revolutionized automatic disease diagnosis and improved healthcare accessibility. This paper introduces an automatic diagnostic system leveraging both Explicit and Implicit Symptoms utilized by the Deep-Q Network Reinforcement Policy. The proposed approach revolves around a dialogue-based system, employing a task-oriented framework that is designed for disease diagnosis. The cornerstone of the used methodology lies in the utilization of reinforcement learning, specifically employing the Deep Q-Network (DQN) policy. The utilized DQN network architecture comprises four linear layers, with the Rectified Linear Units (ReLUs) activation function. We conducted a comparative analysis between two deep learning techniques: the DQN and the Policy Gradients (PG). The experimental data was obtained from the extensive Medical Diagnosis Dialogue repository that encompasses 2,374 dialogues, 12 distinct disease categories, and 118 types of symptoms.