Analyzing Shape Parameterization of SD-OCT Optic Nerve Head Images in High Myopes as a Predictor of Visual Field Defects

Abstract

 

Purpose

 

To see if global shape variations in the vitreo-retinal interface (VRI) of myopic eyes with tilted discs as imaged by OCT at the optic nerve head (ONH) can be used to predict the presence of visual field defects in that eye.

 

Methods

 

A total of 94 eyes with high myopia, defined as spherical equivalent refraction < -5 D, were scanned using the optic disc protocol with a Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, CA) at the Chinese University of Hong Kong Department of Ophthalmology. Of these, 61 eyes had associated field defects as determined by the Humphrey Visual Field Analyzer 24-2 and the rest were normal. For each optic disc scan, the VRI was hand-segmented by a trained expert and the results were used to build an active shape model of the surface for those with and without field defects. The shape model was used to perform dimensionality reduction on the entire dataset and to project each hand-segmented VRI into a 15-dimensional, shape-based subspace. A support vector machine (SVM) was then trained on this new, reduced dataset and validated with leave-one-out cross validation. Accompanying figures illustrate modes of variation in our shape model.

 

Results

 

The sensitivity of the SVM for determining whether or not a test shape was indicative of field defects was 85.3% with a specificity of 70% corresponding to a positive predictive value of 83.9% and a negative predictive value of 71.9%.

 

Conclusions

 

Shape analysis is a new potential method to quantitatively analyze an SD-OCT optic disc scan. This can be particularly useful in the setting of high myopia when other parameters such as retinal nerve fiber layer and ganglion cell analysis may be less reliable. Having shown the utility of the shape model for prediction and analysis of disease, further study is warranted. In particular, we plan to investigate the modes of variation of the shape model which best characterize the differences in shape between the eyes with field defects and those without to help determine the presence and degree of disease.

 

 

Illustration of the principal mode of variation in the shape model. The middle image is the mean shape of the training set. The top and bottom images represents this mode plotted at plus and minus three standard deviations from the mean, respectively.

 

 

Same as Figure 1, but depicting the second most prominent mode of variation.

 

Keywords: 550 imaging/image analysis: clinical • 629 optic nerve • 758 visual fields