Scene Reconstruction in Endoscopic Sinus Surgery: A Tool for Detecting Real-Time Changes

Sunday, September 29, 2024

10:18 AM – 10:24 AM EDT

Location: MBCC, 230

Disclosure(s):

Jeremy S. Ruthberg, MD:

faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

Nicole M. Gunderson, BS:

faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

Waleed M. Abuzeid, MD:

faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

Introduction: Current navigation systems in endoscopic sinus surgery are inherently imprecise by design as the initial preoperative reference frame becomes outdated with each surgical modification. Neural Radiance Fields (NeRF) is a novel technique in 3D reconstruction that is capable of providing the framework for a continuously self-updating navigation system while concurrently reducing setup time and hardware required with current image guidance systems.

Methods: Bilateral nasal endoscopy with a 0-degree 4 mm rigid endoscope was performed on a series of 3 cadaver heads (n = 6 nasal cavities total). A HD Karl Storz Telepack was utilized to record endoscopic video data. Optical calibration was performed with endoscopic capture of a chessboard target and known reference frame prior to nasal endoscopy. After endoscopy, Neural Radiance Field (NeRF) computer vision techniques were applied to recorded endoscopic video data to generate a 3D model of each nasal cavity. Rendering times for each 3D reconstruction were recorded. Preliminary analyses of accuracy measures were performed on a high resolution 3D printed model with varying target sizes (1-2 mm). Target dimensions were validated by micro CT (slice thickness: 12 micron resolution).

Results: 3D reconstruction rendering times ranged from t = 20-21.8 s after the initial training pipeline. Discrepancies between NeRF 3D reconstruction and our validated 3D model ranged from 0.03 mm to 0.26 mm. Comprehensive volumetric analyses of 3D reconstruction to cadaver nasal cavity are ongoing.

Conclusions: NeRF offers rapid rendering and a higher level of precision compared to current navigation systems that operate with 1-2 mm error margins. Novel breakthroughs in computer vision technology may address current, deficient navigation systems in endoscopic sinus surgery, providing updates to the initial reference frame in real-time, with each surgical modification performed. With enhanced computing power, NeRF will be capable of near-instantaneous rendering speeds far improved from our preliminary analyses.

Presenting Author(s)

Jeremy S. Ruthberg, MD

Resident
University of Washington Department of Otolaryngology
Seattle, WA, United States

Disclosure(s):
faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

Co-Author(s)

NG

Nicole M. Gunderson, BS

PhD Student
University of Washington
Seattle, WA, United States

Disclosure(s):
faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.
KM

Kris Moe, MD

Professor
University of Washington
Seattle, Washington, USA

Disclosure information not submitted.
ES

Eric J. Seibel, PhD

Research Professor
University of Washington
Seattle, Washington, United States

Disclosure information not submitted.
WA

Waleed M. Abuzeid, MD

Associate Professor
University of Washington
Seattle, WA, United States

Disclosure(s):
faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

Senior Author(s)

Randall A. Bly, MD

Associate Professor
Seattle Childrens Hospital
Seattle, WA, United States

Disclosure information not submitted.