Single Frame Virtual Averaging Improved Optical Coherence Tomography (OCT) Images Independent from Eye Movement Artifacts

Abstract

 

Purpose

 

Signal averaging is most effective in OCT image enhancement when a built-in eye-tracker is present. For devices without eye-trackers, virtually mimicking this averaging procedure was previously proven to be successful. This approach assumed good spatial integrity within a given three-dimensional cube scan. Hence eye movements during scanning (e.g. saccade), which is predominant in diseased eyes, critically affected its performance due to its neighboring frame referencing approach (Figure). The purpose of this study was to develop and test a single frame virtual averaging approach.

 

Methods

 

Ten eyes of 10 healthy volunteers were scanned with a spectral domain OCT device (Cirrus HD-OCT; Zeiss, Dublin, CA; 200x200 Macular Cube Scan). First, the OCT raw data were oversampled transversely on each frame to have 400 A-scan lines per frame. Then nine consecutive A-scan lines contributed to generate an averaged center A-scan with chance proportions that comprise of a uniform distribution. Two output averaged frames were generated; one with sharper borders and another with strong smoothing. Finally, these two frame data were blended in different proportions to obtain more homogenous texture. Quality index (QI) was computed for quantitative image quality assessment, which is a universal image quality parameter for OCT images analogous to signal to noise ratio.

 

Results

 

Subjective image quality assessment of all the single frame virtual averaged images showed notable improvement when compared to the original device output images with clear retinal layer structures while preserving sharp edges (Figure). The mean quality index significantly improved after virtual averaging (69.42 vs. 79.86 [original vs. processed], p<0.0001, paired t-test).

 

Conclusions

 

The proposed single frame virtual averaging method successfully improved SD-OCT image quality to provide better retinal layer visualization than the original image especially on images with notable eye movement artifacts. This technique may provide an alternative way to improve OCT images without the need for signal averaging or eye tracking during image acquisition.