Advanced Dual-Branch U-Net Decoder for Precise and Robust Surgical Instrument and Organ Segmentation

摘要

The accurate segmentation of surgical instruments and organs is crucial for improving the safety, precision, and efficiency of laparoscopic and robot-assisted surgeries, where real-time guidance and error reduction are paramount. Surgical environments present significant challenges, such as image noise, organ deformation, and tool occlusion, making precise segmentation difficult. This paper presents a surgical image segmentation approach based on a modified U-Net model. The encoder of the neural network used residual blocks, and the decoder architecture is based on a dual-branch model to improve the segmentation performance. One branch focuses on segmenting surgical instruments, while the other on the intervened organ. This design ensures the detection of specific features from the surgical instruments and organs, improving segmentation accuracy. Experimental results evaluate the performance of the proposed neural network model using real surgical datasets. The proposed method achieved superior results, with a mean Intersection over Union (IoU) of 92% for surgical instruments segmentation and 96% for organ segmentation. These findings demonstrate the robustness of the proposed model in challenging surgical environments.