Evaluation of Inner to Outer Retina Area Ratio for Detection of Glaucoma with Macular OCT Imaging

Abstract

 

Purpose

 

To determine whether the inner to outer retina (OR) area ratio can improve detection of glaucoma with macular SD-OCT imaging as compared to ganglion cell/inner plexiform layer (GC/IPL) thickness measurements.

 

Methods

 

Forty-one normal subjects and 27 glaucoma patients from the UCLA OCT Imaging Study were included. Macular images were obtained (posterior pole algorithm, Spectralis SD-OCT) and the raw data files were exported for automatic segmentation of retinal layers using 3-D contextual information (Iowa OCT Explorer). Properties of the derivative of surface and thickness functions were exploited to locate the foveal center in MATLAB. The B-scan passing through the foveal center was used for computing the total area of inner and outer retina layers in the central 4.8mm of the scan. Average ganglion cell/inner plexiform layer (GC/IPL) measurements were also obtained (Macular Cube 200x200, Cirrus HD-OCT). Logistic regression and area under the ROC curves (AUCs) were used to compare models. Performance of ganglion cell complex and GC/IPL to OR (GCC/OR vs. GC/OR) area ratios were also compared.

 

Results

 

The median (IQR) MD was −3.9 (−6.3 to −2.5) dB and 0 (−0.8 to 0.4) dB in glaucoma and normal subjects, respectively. The corresponding mean GC/OR area ratios were 0.27 (±0.05) and 0.33 (±0.03). GC/OR area ratio was not related to age or axial length in normal subjects (p>0.05). GC/IPL thickness and GC/OR ratio were highly correlated (p<0.001) in normal (R2=0.39) and glaucoma subjects (R2=0.74). AUCs for detection of glaucoma were similar for GC/IPL (0.90) or GC/OR area ratio (0.88) when adjusted for age and axial length. GC/OR area ratio and its interaction with average GC/IPL were found to be significantly predictive of glaucoma (p=0.005 and 0.009, respectively) when added to GC/IPL. This improved the predictive performance of the logistic regression compared to GC/IPL alone (AUC for the combined model=0.93; p=0.09 for comparison to GC/IPL model). The results were similar when GCC/OR area ratio was used.

 

Conclusions

 

GC/OR area ratio does not vary as a function of age or axial length in normal subjects. Adding GC/OR area ratio to GC/IPL thickness in multivariate predictive models can potentially improve detection of glaucoma.

 

 

Correlation of the macular GC/OR area ratio with average GC/IPL thickness.

 

Keywords: 550 imaging/image analysis: clinical • 585 macula/fovea • 496 detection