High-Fidelity 3D Reconstruction for Accurate Anatomical Measurements in Endoscopic Sinus Surgery

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Computational Anatomy

Background:

• Accurate intraoperative surgical scene representation is vital for precise navigation in endoscopic sinus surgery (ESS).
• Preoperative CT scans become outdated during surgery due to tissue manipulation, limiting their utility.
• Existing endoscopic 3D reconstruction methods lack the required submillimeter accuracy for ESS.

Purpose of the Study:

• To develop a highly accurate intraoperative 3D surgical scene modeling system for ESS.
• To generate 3D sinus reconstructions that provide real-time anatomical information diverging from preoperative CT.
• To improve surgical navigation and safety by overcoming limitations of outdated preoperative imaging.

Main Methods:

• An expanded Neural Radiance Fields (NeRF) pipeline was developed for monocular endoscopic video.
• Methods to simulate stereoscopic views and iteratively refine reconstruction depth were incorporated.
• Point cloud denoising, outlier removal, and dropout patching were implemented for robustness.

Main Results:

• The system generated high-resolution and accurate 3D reconstructions of the surgical scene without external tracking.
• Accurate depth maps, global scaling, and geometric information were obtained.
• Evaluation on cadaveric specimens showed average reconstruction errors of 0.25mm for ethmoid length and 0.52mm for height.

Conclusions:

• The proposed NeRF-based workflow enables accurate intraoperative 3D modeling for endoscopic sinus surgery.
• This technology offers a robust alternative to outdated preoperative imaging for surgical planning and navigation.
• The achieved submillimeter accuracy supports safer and more precise interventions in complex sinonasal anatomy.