Enhancing Surgical Precision: Deep Learning-Based Depth Estimation in Minimally Invasive Surgery with the MiDaS Model

Minimally Invasive Surgery (MIS) has revolutionised surgical procedures, offering patients less invasive and more efficient treatments. However, MIS presents limited view and depth perception challenges, impacting surgical accuracy and safety. This study focuses on advancing depth estimation in MIS, exploring a range of methodologies to enhance precision and efficiency. We conduct an exhaustive review of contemporary approaches, encompassing conventional methods like stereo matching and structure from motion alongside cutting-edge deep learning techniques. We address specific challenges MIS poses, including issues related to low image quality and the non-rigid nature of tissues. We introduce an innovative deep learning-based framework, leveraging the MiDaS model for depth estimation of endoscopic images. This framework employs convolutional neural networks (CNN) to map input images to their corresponding depth maps. In conclusion, we envision a multitude of potential applications and future directions for depth estimation within MIS, emphasising its potential to enhance surgical precision and safety.