Abstract
Purpose: :
Extracting quantitative information from OCT images requires the identification of boundaries between structures. This segmentation process ideally should be carried out automatically by robust computer algorithms able to handle the variety of pathologies seen in clinical practice. Automatic segmentation is particularly important for the very large three–dimensional (3–D) data sets produced by spectral–domain OCT. This study investigated the boundaries present at the inner retinal surface in patients with epiretinal membranes (ERMs).
Methods: :
Spectral–domain OCT (8 µm axial resolution) was used to obtain 3–D data sets representing a 6x6 mm region of the macula in patients with ERMs. A custom–designed image viewer produced cross–sectional OCT images (B–scans) from the 3–D data and allowed both automated and manual drawing of retinal boundaries. Boundaries from B–scans were assembled to display the 2–D retinal surface in three dimensional space.
Results: :
In B–scans an ERM appeared as a thin, smooth hyper–reflective band between the neural retina and the vitreous. The ERM could lie immediately adjacent to the retina, in which case a boundary that followed the ERM also followed the retinal surface. More frequently, the ERM and retina were adherent at some places along a B–scan and at other places were separated by hypo–reflective spaces. In these cases, when assembled as surfaces in space, boundaries that followed the ERM formed smooth sheets, whereas boundaries that followed the neural retina showed a series of folds or wrinkles that revealed the tension placed on the retina by the ERM. Retinal thickness calculated from an ERM boundary is necessarily larger than when calculated from the retinal surface, but the difference is probably not clinically significant in most cases.
Conclusions: :
Segmentation of OCT images, i.e., the identification of boundaries between structures, is essential for extracting quantitative information. The presence of an ERM can complicate the selection of the inner retinal boundary. Being able to differentiate between these surfaces may not be necessary for determining retinal thickness, but it could offer new insights into interesting aspects of some retinal pathologies.
Keywords: imaging/image analysis: clinical • proliferative vitreoretinopathy • macula/fovea