Weakly Supervised Pose Estimation of Surgical Instrument from a Single Endoscopic Image

Area of Science:

• Computer-aided surgery
• Medical imaging
• Robotics

Background:

• Accurate instrument pose estimation is crucial for computer-aided surgery.
• Challenges include unstable feature points due to refraction and complex backgrounds, and a lack of labeled pose data.

Purpose of the Study:

• To develop a weakly supervised method for surgical instrument pose estimation using single endoscopic images.
• To overcome limitations of existing methods in challenging surgical environments.

Main Methods:

• A three-module approach: segmentation for instrument detection, point inference for feature point prediction, and a Perspective-n-Point module for pose estimation.
• Utilizes instrument image segmentation contours and synthesized endoscopic images.
• Minimizes local feature matching errors and global contour inconsistencies.

Main Results:

• The proposed method demonstrates effective pose estimation for surgical instruments in endoscopic systems.
• Validation performed on both real and synthetic endoscopic images.
• Outperforms current state-of-the-art methods in accuracy and robustness.

Conclusions:

• The developed weakly supervised method offers a viable solution for surgical instrument pose estimation.
• The approach effectively handles challenges posed by endoscopic imaging conditions.
• Contributes to advancing the capabilities of computer-aided surgical navigation.