Original Article
Automated registration and enhanced processing of clinical optical coherence tomography angiography
Abstract
Background: Motion artifacts degrade the quality of optical coherence tomography angiography (OCTA). Orthogonal registration can eliminate the majority of these artifacts, but some artifacts persist in most clinical images. We evaluate an automated registration algorithm with selective merging and filtering to remove remaining artifacts and improve the quality of images.
Methods: A 70 kHz commercial spectral domain OCT was used to obtain 3 mm × 3 mm OCTA in 10 healthy, 5 age-related macular degeneration (AMD), and 31 diabetic retinopathy (DR) participants. Projection artifacts were removed and images were segmented into 3 inner retinal plexuses. Amplitude thresholding identified lines containing a residual artifact and correlation between neighboring lines identified distorted stripes. Then the angiograms were registered and the lines selectively merged. A vesselness filter was applied to the resulting images. The images were evaluated for signal-to-noise ratio (SNR), image entropy, vessel connectivity and vessel density.
Results: Registration and selective merging (RSM) algorithm improved the SNR (P<0.02) compared to orthogonal registration alone. RSM with vesselness filter increased the image entropy (P<10−8) and reduced inter-subject variability (standard error ≤3%, n=10) in healthy eyes. The method improved vessel details and connectivity in OCTA of healthy, DR and neovascular AMD eyes.
Conclusions: This automated registration method eliminates residual motion artifacts and enhances the visualization of vessels in OCTA.
Methods: A 70 kHz commercial spectral domain OCT was used to obtain 3 mm × 3 mm OCTA in 10 healthy, 5 age-related macular degeneration (AMD), and 31 diabetic retinopathy (DR) participants. Projection artifacts were removed and images were segmented into 3 inner retinal plexuses. Amplitude thresholding identified lines containing a residual artifact and correlation between neighboring lines identified distorted stripes. Then the angiograms were registered and the lines selectively merged. A vesselness filter was applied to the resulting images. The images were evaluated for signal-to-noise ratio (SNR), image entropy, vessel connectivity and vessel density.
Results: Registration and selective merging (RSM) algorithm improved the SNR (P<0.02) compared to orthogonal registration alone. RSM with vesselness filter increased the image entropy (P<10−8) and reduced inter-subject variability (standard error ≤3%, n=10) in healthy eyes. The method improved vessel details and connectivity in OCTA of healthy, DR and neovascular AMD eyes.
Conclusions: This automated registration method eliminates residual motion artifacts and enhances the visualization of vessels in OCTA.
