June 2015
Volume 56, Issue 7
Free
ARVO Annual Meeting Abstract  |   June 2015
The effect of manually refining automated image segmentations on retinal nerve fiber layer thickness measurements from optical coherence tomography (OCT) scans
Author Affiliations & Notes
  • Shivali Menda
    Department of Ophthalmology, Devers Eye Institute, Portland, OR
  • Brad Fortune
    Department of Ophthalmology, Devers Eye Institute, Portland, OR
  • Steven L Mansberger
    Department of Ophthalmology, Devers Eye Institute, Portland, OR
  • Stuart Keith Gardiner
    Department of Ophthalmology, Devers Eye Institute, Portland, OR
  • Shaban Demirel
    Department of Ophthalmology, Devers Eye Institute, Portland, OR
  • Footnotes
    Commercial Relationships Shivali Menda, None; Brad Fortune, None; Steven Mansberger, Alcon (C), Allergan (C), Allergan (S), Envisia (C), Forsight Vision5 (C), Mobius (S), NEI (S), Santen (C), Welch Allyn (C); Stuart Gardiner, Carl Zeiss (C); Shaban Demirel, Carl Zeiss Meditec (F), Heidelberg Engineering (F)
  • Footnotes
    Support None
Investigative Ophthalmology & Visual Science June 2015, Vol.56, 4551. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Shivali Menda, Brad Fortune, Steven L Mansberger, Stuart Keith Gardiner, Shaban Demirel; The effect of manually refining automated image segmentations on retinal nerve fiber layer thickness measurements from optical coherence tomography (OCT) scans. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):4551.

      Download citation file:


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

      ×
  • Supplements
Abstract
 
Purpose
 

OCT is helpful for monitoring progressive glaucomatous loss of retinal nerve fiber layer thickness (RNFLT). However, automated image segmentation algorithms used to determine the boundaries of the RNFL occasionally fail. This study investigates the effect of experienced operators manually refining these boundaries on RNFLT measurements.

 
Methods
 

We used Spectralis (Heidelberg Engineering) OCT peripapillary circle scan data from 23 participants in the Portland Progression Project. For this study, each individual was scanned 5 times within ≈ 2 months (12 deg diameter circular B-scan with 1536 A-lines. Nine sweeps averaged for the final b-scan). For each scan, we extracted the 1536 RNFLT samples for the instrument’s native automated segmentations. Then experienced operators manually refined the segmentations for the anterior (internal limiting membrane) and posterior RNFL boundaries to correct obvious errors. We compared manual refinement to automated RNFLT for differences in global RNFLT and sectoral RNFLT (12 clock-hours).

 
Results
 

We excluded both eyes of 1 individual because of vitreo-retinal traction and RNFL distortion. For the remaining 44 eyes, manual refinement increased the global average RNFLT by only 0.43μm (P=0.469; mixed effects models). However, we found large differences in the effect of manual refinement between eyes (95% confidence interval -10.59μm to +10.60μm) that were approximately twice the range of test-retest variability. The largest difference in global RNFLT was -19.4μm. An example of segmentation disagreement is shown in Figure 1. Sectors in the superior RNFL showed the largest differences between automated and manually refined segmentations (See Figure 2. 11, 12 and 1 o’clock sectors; P=0.003, 0.002 and 0.003 respectively).

 
Conclusions
 

Manual refinement of automated RNFL segmentation can correct large errors, increasing the possibility of detecting clinically meaningful differences of RNFLT measurements for some eyes. The largest errors in automated segmentation tended to occur in regions of the RNFL that are important for glaucoma management. Researchers and clinicians should inspect OCT scans and perform manual refinement of segmentations to correctly estimate RNFLT.  

 

 
×
×

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.

×