Global region reidentification for camera relocalization in video-based surgical navigation.

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Navigation

Background:

• Vision-based navigation systems require accurate camera poses for surgical guidance.
• Endoscopic camera registration degrades upon reinsertion, necessitating relocalization.
• Accurate reidentification of the surgical scene and camera pose estimation are challenging.

Purpose of the Study:

• To develop a training-free method for endoscopic camera relocalization after reinsertion.
• To accurately reidentify the region of interest (ROI) and estimate camera pose.
• To enable pose reinitialization for vision-based surgical applications.

Main Methods:

• Utilized previously observed images with known poses and a CT scan.
• Combined foundation models with classical techniques for global reidentification of the ROI.
• Employed image-based feature matching and Perspective-n-Point (PnP) algorithm for pose recovery.

Main Results:

• Experiments conducted on eight sequences from three cadaver studies.
• Method accurately reidentifies endoscope entry into the ROI and selects suitable image pairs for PnP.
• Achieved average translation error of 1.74 mm and rotational error of 0.09 radians.

Conclusions:

• Presents a training-free approach for detecting endoscope re-entry into the ROI and estimating camera pose.
• Demonstrates promising results for pose reinitialization in vision-based surgical navigation.
• Enables automated, intervention-free reinitialization for enhanced surgical guidance.