Purchase this article with an account.
Acner Camino, Yali Jia, Jie Wang, Qisheng You, Xiang Wei, Liang Liu, David Huang; Automated detection of shadow artifacts in OCTA. Invest. Ophthalmol. Vis. Sci. 2019;60(11):PB055. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Artifacts in optical coherence tomography angiography (OCTA) can be caused by motion, projections of superficial blood flow or shadows from opaque objects anterior to the retina (e.g. vitreous floaters, pupil boundary). While motion and projection artifacts have been studied extensively, shadow artifacts have been overlooked thus far. We aim at detecting shadow artifacts to exclude them from quantitative OCTA metrics.
3×3 and 6×6 mm2 macular OCTA scans were acquired by a spectral-domain OCT system (AngioVue). Retinal layers were segmented by a directional graph search algorithm implemented in a customized image processing platform. Background signal from avascular tissue was removed by an iterative, regression based bulk motion subtraction algorithm. A machine learning ensemble method was trained with OCTA of healthy subjects where shadows were manufactured by placing a PLA filament between the cornea and the instrument. Labeling of the shadow positions used in training was based on variation of the local vessel density compared to a scan of the same subject under optimal imaging conditions. The features used for shadow classification were local flow index, low projected reflectance in both the inner and outer retina, and low standard deviation of the reflectance in the retinal slab. Software performance was evaluated in scans acquired from healthy subjects by progressively increasing signal attenuation with neutral density filters (NDF).
The sensitivity and specificity of the software classification were 91.6% and 86.9%, taking the manual segmentation of one expert grader as ground truth. Another expert grader performed with a sensitivity of 87.2% and a specificity of 93.3% with respect to the same reference. Shadows were detected on OCTA scans of intermediate uveitis and diabetic retinopathy (Fig.1). Vessel density of participants in the NDF experiment was independent of signal strength after shadow exclusion (Fig. 2).
Areas of unreliable OCTA signal due to shadows were successfully detected in diabetic retinopathy and uveitis. As OCTA becomes a common imaging modality used in the ophthalmic practice, exclusion of these artifacts will help to improve the confidence with which OCTA is used to diagnose and evaluate retinal diseases. This is important as older patients often have cataracts, vitreous opacities, small pupil, dry eye – conditions in which shadows are difficult to prevent during imaging sessions.
This abstract was presented at the 2019 ARVO Imaging in the Eye Conference, held in Vancouver, Canada, April 26-27, 2019.
This PDF is available to Subscribers Only