June 2013
Volume 54, Issue 15
Free
ARVO Annual Meeting Abstract  |   June 2013
The Prevalence of Cirrus SD-OCT Ganglion Cell Segmentation Errors in High Myopes
Author Affiliations & Notes
  • Tessa Johung
    Stanford University, Stanford, CA
  • Jonathan Oakley
    Voxeleron, Pleasanton, CA
  • Daniel Russakoff
    Voxeleron, Pleasanton, CA
  • Sonny Sabhlok
    Stanford University, Stanford, CA
  • Felix Li
    Dept. of Ophthalmology and Visual Sciences, Chinese University of Hong Kong, Hong Kong, China
  • Robert Chang
    Stanford University, Stanford, CA
  • Footnotes
    Commercial Relationships Tessa Johung, None; Jonathan Oakley, Voxeleron LLC (E); Daniel Russakoff, Voxeleron (E); Sonny Sabhlok, None; Felix Li, None; Robert Chang, None
  • Footnotes
    Support None
Investigative Ophthalmology & Visual Science June 2013, Vol.54, 4845. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Tessa Johung, Jonathan Oakley, Daniel Russakoff, Sonny Sabhlok, Felix Li, Robert Chang; The Prevalence of Cirrus SD-OCT Ganglion Cell Segmentation Errors in High Myopes. Invest. Ophthalmol. Vis. Sci. 2013;54(15):4845.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract
 
Purpose
 

To investigate the prevalence of sectoral value differences due to segmentation errors in macular Ganglion cell-inner plexiform layer (GCIPL) scans using SD-OCT in glaucomatous and highly myopic eyes.

 
Methods
 

174 Cirrus HD-OCT macula scans were obtained from 77 patients. 96 eyes had high myopia (> -5D acuity) and were glaucoma suspects. 78 were defined as glaucomatous, having progressive visual field defects in more than one examination. GCIPL analysis was performed using the Cirrus 6.0 software and custom Septem retinal segmentation software. All thickness measurements were compared and sectors differing by more than 10 microns were manually reviewed by an expert observer. Cirrus review was performed on the instrument; Septem review using a Matlab GUI. A sector error was counted whenever either an error of more than 10 microns was seen.

 
Results
 

For Cirrus, 27 eyes (15.5%) reported at least one sector error. The Septem algorithm had only 6 such instances (3.4%). The chi-square test reports a significant difference in the prevalence of errors between the algorithms (p<0.001). Figure 1 shows example GCIP maps. Figure 2 shows their underlying segmentations.

 
Conclusions
 

The prevalence of errors in GCIP Cirrus thickness maps in the study is high (15.5%). Algorithms are improving, as is evidenced by the low error rate using Septem (3.4%). For current commercial devices, the emphasis on clinical judgment when assessing glaucomatous damage in such populations remains.

 
 
Figure 1: On the left is the GCIPL thickness map from the Cirrus 6.0 software. On the right that generated by Septem. Each colorbar reports thicknesses in microns. Note that the Cirrus map shows two apparent arcuate defects, but each should be turning toward the optic nerve, on the right of the image, to be real (this eye is OD). The Septem map shows no such defects.
 
Figure 1: On the left is the GCIPL thickness map from the Cirrus 6.0 software. On the right that generated by Septem. Each colorbar reports thicknesses in microns. Note that the Cirrus map shows two apparent arcuate defects, but each should be turning toward the optic nerve, on the right of the image, to be real (this eye is OD). The Septem map shows no such defects.
 
 
Figure 2: The top image shows a vertical B-scan from the OCT and Cirrus segmentations. It is positioned to pass through an apparent arcuate defect (Figure 1). The left side is inferior; the right side superior. Below, is the result from Septem, here showing all seven segmented layers. Each has the posterior of the RNFL and IPL reported in magenta and yellow, respectively. The error in the Cirrus result appears to stem from the posterior of the IPL boundary identifying instead the posterior of the RNFL. Arrows show show where the RNFL (magenta arrows) and IPL (yellow arrows) should be.
 
Figure 2: The top image shows a vertical B-scan from the OCT and Cirrus segmentations. It is positioned to pass through an apparent arcuate defect (Figure 1). The left side is inferior; the right side superior. Below, is the result from Septem, here showing all seven segmented layers. Each has the posterior of the RNFL and IPL reported in magenta and yellow, respectively. The error in the Cirrus result appears to stem from the posterior of the IPL boundary identifying instead the posterior of the RNFL. Arrows show show where the RNFL (magenta arrows) and IPL (yellow arrows) should be.
 
Keywords: 549 image processing • 550 imaging/image analysis: clinical • 605 myopia  
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×