Figure 1.

Impact of deep learning denoising on the quality of optical coherence tomography angiography (OCTA) images. (A–C), OCTA images of normal superficial capillary plexus obtained using a 3 × 3 mm² scan pattern. (D–F), Magnified images of open squares in A–C. (G–I), OCTA images of branch retinal vein occlusion. A, D, G, Original OCTA images. B, E, H, Averaged OCTA images. C, F, I, Denoised OCTA images. Averaging and deep learning denoising provide background noise reduction and annealing of disconnected vessels, resulting in noiseless, smooth, and high-contrast images. The contrast-to-noise ratios (CNRs) of A, B, C, G, H, and I are 1.03, 1.42, 1.74, 0.77, 1.16, and 1.27, respectively.