Content area

Abstract

INTRODUCTION

A major obstacle to improving procedure safety and accuracy with image guidance technology is the need for rapid deployment of real-time registration and tracking of a moving patient. An example of this in neurosurgery is the persistence of freehand placement for external ventricular drains (EVDs), which has an inherent risk of inaccurate positioning, multiple passes, tract hemorrhage, injury to adjacent brain parenchyma, catheter occlusion, and infection.

METHODS

Advanced image localization methods were used to develop an algorithm that performs near continuous, automatic, and markerless image registration. This ‘snap-surface alignment’, a proprietary term for this process, aligned computed tomography scans of three human cadaver heads to their respective 3D camera images. The accuracy, speed and root mean square distance produced by the algorithm was recorded and depicted via accuracy heatmaps. The algorithm was evaluated under several test conditions, such as extreme camera angles, surgical draping with limited exposed surface area and facial features, and differential subject lighting intensity.

RESULTS

Registration was successful for all three cadaveric specimens, with a median registration accuracy of 0.86mm (IQR = 1.10mm). Areas of high registration quality included the forehead, zygoma and mental region, with a median accuracy of 0.76mm (IQR = 0.91mm). The median speed of registration was 1.10s (IQR = 0.67s). Image registration was successful in variable test conditions using multiple camera angles (median error = 1.00mm, IQR = 1.22mm), surgical draping (median error = 1.05mm, IQR = 1.32mm) and intense (135 watt) light (1.02mm, IQR = 1.24mm).

CONCLUSION

This computer vision-based registration provided real-time tracking of cadaveric heads with recalibration time of approximately one second with sub-millimetric accuracy. Using this approach to guide bedside ventriculostomy could reduce complications, improve safety, and could be extrapolated to other frameless stereotactic applications.

Details

Title
Computer Vision Registration as a Novel and Accurate Approach for Frameless Stereotactic Neuronavigation
Author
Robertson, Faith C; Sha, Raahil; Amich, Jose; Lal, Avinash; Lee, Benjamin; Wu, Kyle; Segar, David J; Calvachi, Paola; Gormley, William
Publication year
2020
Publication date
Dec 2020
Publisher
Wolters Kluwer Health, Inc.
ISSN
0148396X
e-ISSN
15244040
Source type
Scholarly Journal
Language of publication
English
ProQuest document ID
2502878772
Copyright
Copyright © 2020 by the Congress of Neurological Surgeons