Three-dimensional cube OCT data were obtained from eyes dilated with tropicamide 1% and phenylephrine 2.5% with the same Cirrus HD-OCT device using the Macular Cube 200×200 scan protocol and processed with a prototype algorithm (prerelease version) that is intended to be incorporated into the Cirrus 6.0 software. This protocol performs 200 horizontal B-scans comprising 200 A-scan per B-scan over 1024 samplings within a cube measuring 6 × 6 × 2 mm (
Fig. 1A) and is designed for retinal topography analysis. Five scans per eye obtained on 5 days within 2 months were used to calculate the intervisit measurement variability. Images with signal strength <6 (per manufacturer recommendation) and those with visible eye motion or blinking artifacts were considered of poor quality and discarded.
The GCA algorithm identifies the outer boundary of the RNFL (using a different method than in the peripapillary scans) and the outer boundary of the IPL. The difference between the RNFL and the IPL outer boundary segmentations yields the combined thickness of the RGC layer and the IPL. In the image data, the boundary between these two layers is anatomically indistinct so that they are difficult to distinguish from each other, but the combined thickness is considered to be indicative of the health of RGCs. The average, minimum (lowest GCIPL thickness over a single meridian crossing the annulus), and sectoral (superotemporal, superior, superonasal, inferonasal, inferior, inferotemporal) thicknesses of the GCIPL are measured in an elliptical annulus (dimensions: vertical inner and outer radius of 0.5 mm and 2.0 mm, horizontal inner and outer radius of 0.6 and 2.4 mm, respectively) around the fovea (
Fig. 1A). The size and shape of the annulus are the result of a preliminary analysis performed using GCIPL maps of 47 healthy eyes. The size of the inner ring was chosen to exclude the area where the GCL is thin and difficult to detect, whereas the dimension of the outer ring was selected to conform closely to the real anatomy of the macular region, where the GCL is thickest in a normal eye. Data were processed with a prototype algorithm that is intended to be incorporated into the Cirrus 6.0 software. The algorithm processes data from either of the 3D volume scans from Cirrus; both scan patterns cover the same physical field of view, namely 6 × 6 × 2 mm, but the image data dimensionality is either 512 × 128 × 1024 or 200 × 200 × 1024. The input image data are initially segmented using the existing Cirrus inner limiting membrane (ILM) and RPE segmentation algorithms to create a region of interest within which lie the intraretinal layers. The algorithm continues in such a hierarchical approach, segmenting first the outer boundary of the outer plexiform layer (OPL), followed by the outer boundary of the IPL, and last the outer boundary of the RNFL. The segmentation procedure operates entirely in three dimensions and uses a graph-based algorithm to identify each layer. Image data are transformed into cost images such that the graph algorithm can find the lowest cost surface. To do this, the input data are initially median filtered to reduce speckle noise. They then create cost images based on directional edge-filtered images that have been enhanced to highlight specific boundary intensity changes using a sigmoid function. These are combined with positional cost images to form a single representation that is partitioned by the graph segmentation algorithm. The segmentation that results is globally optimal in terms of its cumulative cost for each of the layers.
The software analyzes the values, compares them to the device's internal normative database, and generates a thickness map, a deviation map, and a significance map color-coded to match RNFL thickness, with values within the normal range in green (P = 5%–95%), borderline values in yellow (1% < P < 5%), and values outside the normal range in red (P < 1%). The normative database contains data from 282 healthy subjects (133 men, 149 women) aged 19 to 84 years (mean age, 48.2 ± 16.9 years). Ethnicity breakdown of the Cirrus RNFL and macula normative databases was as follows: 43% Caucasian, 24% Asian, 18% African American, 12% Hispanic, 1% Indian, and 6% mixed ethnicity. Mean values were as follows: axial length, 23.94 ± 1.06 mm (range, 20–28 mm); IOP, 14.0 ± 2.4 mm Hg (range, 8–20 mm Hg); refraction expressed as spherical equivalent, −0.9 ± 2.1 D (range, −10.6 to +6.25 D); central corneal thickness (CCT), 549.58 ± 36.67 μm (mean, 449–662 μm); and VF field mean deviation, 0.02 ± 1.04 dB (range, −2.07 to +2.58 dB). All ILM and RNFL scans and those posterior to the IPL and OPL boundaries were manually reviewed for accuracy. The segmentation result was overlaid onto the OCT volume data and visualized in computing software (MatLab; MathWorks, Natick, MA), which often allows visualizing algorithm errors in the en face thickness maps. Any scan with an apparent segmentation error detected during this process was excluded from the study. More specifically, each B-scan was reviewed to check that the segmentation error of the ILM, OPL, or inner segment/outer segment boundary was <10 pixels for >75% of all A-scans within the annulus region. Any scan that did not meet this criterion was excluded.