Optimizing Retinal Surgery: Reinforcement Learning for Enhanced Microscope-Assisted Robotics

Abstract

Recent progress in ophthalmology provides advanced operating rooms with surgical robots and microscopes. Integrating these tools has a significant impact on the field of retinal surgery. Traditional retinal surgeries were often limited by the risk of tremors and challenges in maintaining steady control during complex surgical procedures, leading to a higher risk of complications. This study proposes a Reinforcement Learning (RL) approach to control a robotic arm in retinal microsurgery to enhance precision and reduce the inherent risks of this delicate procedure. The proposed model consists of several key elements, such as a robotic surgery arm, a microscope, and RL agents to control the surgical instrument in real-time according to the visual feedback from the microscope. The RL agent employs a Deep Q-Network (DQN) architecture by interacting with the environment through a sequence of actions and rewards to enhance the movement of the robotic arm. The model utilizes a Convolutional Neural Network (CNN) to extract features from images or frames for accurate state representation. The results demonstrated superior performance with an accuracy of 95%, precision of 97%, recall of 96%, and an F1 score of 96%. The simulation results confirm the high precision control of the robotic arm for minimizing complications in retinal surgeries.