Sonya B. Shah, MD; L. Jay Katz, MD

Disclosures

June 07, 2012

Color Reflectivity Discretization Analysis (CORDA) of Optical Coherence Tomography Images in the Detection of Glaucomatous Retinal Nerve Fiber Layer Defects

Optical coherence tomography (OCT) has become a widely used ocular imaging modality. The recent introduction of spectral domain OCT has vastly improved imaging speed, resolution, and quality over time domain technology.

Cirrus™ HD-OCT (Carl Zeiss Meditec; Dublin, California) obtains a spectral domain image of the retinal nerve fiber layer (RNFL) and is useful for evaluating the optic nerve head. The Optic Disc Cube 200x200 program obtains 200 A-scans from 200 linear B-scans across a 6-mm2 area centered on the optic nerve head.[1]

The thickness of the RNFL is then analyzed. Mean and sectoral RNFL thicknesses are compared across an internal normative database. Those sectors are marked as abnormal (outside of the 99% confidence interval) or borderline (95%-99% confidence interval) compared with age-matched normals.

Color reflectivity discretization analysis (CORDATM, Diopsys®, Inc; Pine Brook, New Jersey) is a new method under development for OCT image analysis. CORDA evaluates reflectivity of the various components of the RNFL on SD-OCT images. Different components of the images are assigned reflectance values, and each of these values corresponds to a color. The colored dots are visible on the final B-scan image. The CORDA software analyzes various colored dots to quantify 4 predefined high-reflectivity (HR) or low-reflectivity (LR) parameters (HR1, HR2, LR1, and LR2).

Hypothesis. It was hypothesized that distinct cellular constituents can be isolated and measured with discretization. This is based on the hypothesis that the HR2 parameter represents the axons of the ganglion cells and the LR parameters represent glial and other supporting tissues. The analysis excludes retinal tissue posterior to the ganglion cell layer to avoid highly reflective areas such as the retinal pigment epithelium.

Methods. All patients were examined by experienced observers in the Glaucoma Department at Wills Eye Institute. Patients underwent slit lamp examination of the optic nerve head and RNFL, Cirrus OCT, and standard automated perimetry.

Three groups of patients were studied. The primary open angle glaucoma (POAG) group consisted of patients with a glaucomatous optic nerve appearance and corresponding visual field defects on 2 or more tests. Angle closure or any secondary glaucoma was excluded. The POAG suspect/ocular hypertension group consisted of individuals with a glaucomatous optic nerve appearance or Goldmann intraocular pressure (IOP) measurements > 22 mm Hg on 2 visits with normal visual fields. Finally, the normal group included those with normal optic nerve appearance and visual field, with IOP < 22 mm Hg.

Mean deviation from Octopus® (Haag-Streit; Koeniz, Switzerland) visual field for patients with glaucoma was obtained. Fields were conducted within 6 months of the OCT images used in this study. Cirrus OCT B-scans were exported as JPEG images from the Cirrus machine and were uploaded into the CORDA software for analysis. Spearman correlations between HR2, RNFL thickness, and mean deviation were calculated.

Results. The total population for the study was 138; 49 with glaucoma, 51 glaucoma suspects, and 38 normal patients were included. The average mean deviation on visual field testing for patients with glaucoma was -9.0 ± 5.1db. Global CORDA parameter HR2 displayed strong correlation with mean RNFL thickness (r = 0.70, P < .001). However, correlations between RNFL thickness and LR1 and LR2 were much lower (r = 0.22, P = .011 for each) (Figure). HR1 was not studied because the very high reflectance values in this category are thought to be a result of optical scatter.

Figure. Mean HR2, LR1, and LR2 and average RNFL thickness, all patients.

Average RNFL thickness and inferior RNFL thickness each correlated well with mean deviation (r = 0.71, P < .001). Superior RNFL thickness also correlated well with mean deviation (r = 0.64, P < .001).

processing....