Laparoscopic surgeries are minimally invasive, requiring only small incisions which result in faster patient recovery and a lower risk of complications. Despite these advantages, surgeons face some challenges, such as limited visibility and control over instruments, potentially compromising precision and coordination during procedures. To address these limitations, advanced technological systems enhance the visibility, control, and overall effectiveness of laparoscopic surgeries.

This research introduces an instrument segmentation method using a modified U-Net model. The model integrates residual blocks in the encoder to optimize learning and prevent gradient degradation, enabling the capture of complex patterns. The decoder is designed with two branches: one focused on instrument segmentation and the other on background segmentation. By combining both outputs, the system improves the accuracy and efficiency of segmenting surgical instruments in real-time.

The system's performance was evaluated through metrics such as the Jaccard index, precision, recall, F1 score, and accuracy. Tests under geometric and signal processing distortions were also conducted to replicate varying surgical conditions, revealing the system's high robustness and adaptability. The results show a high efficiency with an accuracy of 0.94 and a Jaccard index of 0.93. Additionally, this approach demonstrates significant improvements in identifying instruments accurately and reducing potential patient injury.

This development enhances surgical precision and increases patient safety during laparoscopic procedures. Furthermore, it provides a valuable tool for training and evaluating surgeons' psychomotor skills. This innovation represents a step toward the future of minimally invasive surgery, minimizing direct surgeon intervention and improving overall patient outcomes.