2.3. Smoothing function definition

The smoothing S(x,y;β)^k is introduced to reduce the staircase effect [45] when the WM filter is repeated a large iterative number of times (due to the nature of the median operation) and greatly improved the quality of the segmentation results. Its definition is based on a similar idea to the “flattening” usually done in SD-OCT segmentation algorithms [19,22,35,46], in which depth distances are considered with respect to estimations of the ILM or RPE layers, and not with respect to the axial axis origin at the top of the complete cube. A “flattening” step relative to the ILM layer usually helps the segmentation of inner retinal layers in healthy eyes, since they tend to follow a similar curvature. However, in eyes with severe abnormalities, the retinal layers can take curvatures very different from the ILM due to, for example, drusen, fluid accumulations, or RPE detachment. In this work, we considered S(x,y;β)^k a smoothed version of the boundary S(x,y)^k, computed using a diffusion regularization fitting operation with a smoothing factor of β. The fitting tools employed in this characterization are available online for download [47].