Monocular Vision-Based Endoscopic Sinus Navigation: A SLAM Driven Approach With CT Integration.

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Otolaryngology

Background:

• Surgical navigation is crucial for precision in sinus surgery, especially near critical structures.
• Existing navigation systems often rely on external trackers, which can be intrusive.
• Monocular endoscopy navigation faces challenges in reconstruction and pose estimation.

Purpose of the Study:

• To present a proof-of-concept for monocular vision-based sinus navigation using only preoperative CT data and endoscope video.
• To develop a system that enhances spatial awareness and precision during sinus surgery.
• To evaluate the accuracy and feasibility of a vision-based navigation system.

Main Methods:

• Developed a vision-based navigation system employing SLAM (Simultaneous Localization and Mapping) for camera pose estimation and 3D surface reconstruction.
• Integrated preoperative CT data with the endoscope video stream.
• Used semi-automated registration and interactive scale solving with reconstructed point clouds.
• Mapped SLAM trajectories to CT space and visualized updates in triplane view.

Main Results:

• Off-site testing on cadaveric data yielded translation errors of 3.2 mm and rotation errors of 4.9 degrees compared to optical tracker data.
• On-site testing was successfully performed in cadaver experiments.
• The system demonstrated a closed loop between image-based reconstruction and CT visualization.

Conclusions:

• The developed system shows promise for monocular vision-based sinus navigation.
• Further research is needed to address challenges for clinical-level surgical navigation.
• This approach offers a less intrusive alternative to traditional navigation systems.