Shape Context Algorithm Applied to Correct Eye Movement Artifacts on Three-Dimensional (3D) Spectral Domain Optical Coherence Tomography (SD-OCT)

Abstract

Purpose: : Eye movement during 3D SD-OCT scanning creates significant distortion in 3D image data that may influence image interpretation/analysis. The purpose of this study was to develop a fully automated system for eye movement correction of 3D SD-OCT images using shape context algorithm aimed for preserving natural optic disc contours and to evaluate its performance with subjective validation.

Methods: : 3D SD-OCT (Cirrus HD-OCT; Carl Zeiss Meditec, Inc., Dublin, CA) images with visible distortion and corresponding scanning laser ophthalmoscopy (SLO) images were included in the study. 3D OCT data were registered to the corresponding SLO images using global transformation computed based on the shape context of detected corresponding points, which were automatically defined based on retinal blood vessel maps detected on both SLO and 3D OCT data. Three observers independently evaluated each processed image with respect to overall fit and disc contour fit. The average opinion served as gold-standard.

Results: : Sixty-seven images (21 healthy and 46 glaucomatous images; 43 eyes (19 healthy and 24 glaucomatous) of 39 subjects (17 healthy and 22 glaucomatous)) were enrolled. The software failed to obtain the final processing results for 3 images. After correction, overall fit was improved in 86.1% of glaucomatous and 95.2% of normal images, equivalent in 13.9% of glaucomatous and 4.8% of normal images, and degraded in 0% of all images. 95.3% of disc contour improved or preserved to original shapes after movement correction.

Conclusions: : Our newly developed fully automated software may be useful for correcting eye movement artifacts in 3D SD-OCT images.

Keywords: image processing • imaging/image analysis: non-clinical • retina