Fig. 1. Flowchart of the methodology illustrating the corresponding main processing steps that need to be performed to automatically extract the cellular layers of the retina on OCT images.

Fig. 2. Optimal choice of the threshold value (κ) and iteration number (N) using small values of θ to obtain the optimal image restoration result. (A) Selection criteria of optimal stopping time. The iteration should be stopped after 40-50 iterations to avoid redundancy computation (κ= 10, with θ = π/30, σ= 1 and Δt=0.24). (B) Measured S/MSE improvement of a typical OCT image as κ is varied from 1 to 20 (N= 50, with θ = π/30,σ= 1 and Δt= 0.24).

Fig. 3. Denoising results for a sample OCT scan. (A) Original OCT image. [Media 1](B) Image denoised (real part) using the nonlinear complex diffusion filter (κ=10, θ=π/30, σ =1, N=50, and Δt=0.24). (C) Imaginary part of the original OCT image after nonlinear complex diffusion filtering (κ = 10, θ = π/30, σ= 1, N=50, and Δt=0.24). (D) Image obtained after a coherence-enhanced diffusion filtering (α=4, σ=5, ρ =2, m=8, and Δt=0.24) is applied to the denoised image (real part shown in B) obtained after nonlinear complex diffusion filtering. (E) Image denoised (real part) using the nonlinear complex diffusion filter (κ= 60, θ= π/30, σ =1, N=50, and Δt=0.24). (F) Imaginary part of the original OCT image after nonlinear complex diffusion filtering (κ = 60, θ = π/30, σ= 1, N=50, and Δt=0.24). (G) Image obtained after a coherence-enhanced diffusion filtering is applied to the denoised image (real part shown in E) obtained after nonlinear complex diffusion filtering. (H) Edge map obtained calculating the first derivative of the structure coherence matrix of the denoised image shown in E. The OCT images displayed are grayscale representations of the actual interference signal intensities.

Fig. 4. Local coherence and orientation in a sample OCT scan. Top: Structure tensor coherence. Bottom: Structure tensor orientation (in degrees).

Fig. 5. Segmentation of an A-scan line based on the coherence structure information extracted from the OCT signal intensity after enhancing diffusion filtering.

Fig. 6. Automated segmentation results. Top: Original OCT image with overlaid retinal boundaries. The segmented retinal layers are, from top to bottom, retinal nerve fiber layer (RNFL), ganglion cell layer (GCL) along with the inner plexiform layer (IPL), inner nuclear layer (INL), outer plexiform layer (OPL), outer nuclear layer (ONL) and the photoreceptor inner/outer segment junction (IS/OS). The retinal pigment epithelium (RPE) along with the choriocapillaries (ChCap) and choroid layer appear below the bottom boundary line. Bottom: A small section of the original OCT image containing 100 A-scans and the single axial scan shown on Fig. 3. The tomogram is composed of 512 A-scans. We note that the sublayer labeled as the ONL is actually enclosing the external limiting membrane (ELM) but in the standard 10-15μm resolution OCT image this thin intraretinal layer cannot be visualized clearly. Thus this layer classification is our assumption and does not reflect the actual anatomic structure.

Fig. 7. Automated and semi-automated segmentation results. The boundaries detected are superimposed on the original OCT image shown in Fig. 3(a). Note that we have assumed a constant thickness for the layers obtained with the semi-automated approach (i.e. for the IS/OS junction and ChCap layer).

Fig. 8. Thickness segmentation mapping obtained for the same normal subject shown in Fig. 1 after automatic segmentation of the retinal layers (see Fig. 4). (a) Whole macular thickness map. (b) RNFL thickness map. (c) GCL + IPL thickness map. (d) INL thickness map. (e) OPL thickness map. (f) ONL thickness map. Note that the scale of the color scheme in the maps is adjusted for the thickness range of each extracted layer. Thickness values are in microns.

Fig. 9. Automatic segmentation results for two pathologic human eyes. Top: Glaucomatous subject. Bottom: Subject with a small subfoveal cyst.

Fig. 10. Automatic segmentation results for two more pathologic human eyes showing a higher perturbation in the retinal structure. Top: Chorioretinitis. Bottom: Macular edema.