Purchase this article with an account.
Hiroshi Ishikawa, Chieh-Li Chen, Gadi Wollstein, Larry Kagemann, Ian Sigal, Richard Bilonick, James Fujimoto, Joel Schuman; Virtual Frame Averaging - A Novel Image Enhancement Method for Three-Dimensional (3D) Optical Coherence Tomography (OCT) Images. Invest. Ophthalmol. Vis. Sci. 2013;54(15):5541.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Signal averaging is one of the most fundamental and effective techniques to improve OCT image quality. But it requires hardware eye-tracking or special registration that makes it difficult to apply post hoc. The purpose of this study was to develop a novel image enhancement method by virtually averaging OCT signal so that the quality of non-averaged non-tracking OCT images became comparable to the one of active tracking frame-averaged OCT images.
Eleven eyes of 11 normal volunteers had macular scans with non-averaged non-tracking 3D OCT (Cirrus HD-OCT; Carl Zeiss Meditec, Dublin, CA). Corresponding active tracking frame-averaged OCT images (Spectralis, Heidelberg Engineering, Germany) were also collected as references. For each sampling voxel, one voxel value was randomly selected from the 3x3 neighborhood voxels using two-dimensional Gaussian random distribution then a random Gaussian deviation was added. This was repeated 15 times for each voxel, and the average of the outcomes replaced the original value. Subjective quality as well as signal to noise ratio (SNR) and contrast to noise ratio (CNR) were assessed to evaluate the image enhancement effect. In addition, distance between the end of visible nasal retinal nerve fiber layer and the foveola (dNFL) was measured to assess the effect on improved retinal layer visibility.
Subjectively all processed images showed notable improvement and bore clear resemblance to active tracking averaged Spectralis images (Figure). The external limiting membrane became clearly visible (red arrowhead) and the contrast between retinal layers became easily distinguishable (yellow arrowhead). The mean SNR and CNR were significantly improved after virtual averaging (SNR: 33.4 vs 47.8 dB, CNR: 4.2 vs 6.4 dB [original vs processed], p<0.0001, paired t-test). dNFL were significantly different before processing (616 [original] vs 453 μm [Spectralis], p=0.002) but not after virtual averaging (466 [processed] vs 453 μm [Spectralis], p=0.68).
The virtual averaging method successfully improved non-averaged non-tracking OCT image quality and made the images comparable to active tracking frame-averaged OCT images. This method may enable detailed retinal structure studies on images which previously fell short on image quality.
This PDF is available to Subscribers Only