June 2015
Volume 56, Issue 7
Free
ARVO Annual Meeting Abstract  |   June 2015
Automated segmentation of macular SD-OCT scans of retinitis pigmentosa patients shows regional patterns of foveal inner retinal thickening that correlate with visual function
Author Affiliations & Notes
  • Andrew Lang
    Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD
  • Aaron Carass
    Computer Science, Johns Hopkins University, Baltimore, MD
  • Ava K Bittner
    College of Optometry, Nova Southeastern University, Ft. Lauderdale, FL
  • Howard S Ying
    Wilmer Eye Institute, Johns Hopkins University School Of Medicine, Baltimore, MD
  • Jerry L Prince
    Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD
  • Footnotes
    Commercial Relationships Andrew Lang, None; Aaron Carass, None; Ava Bittner, None; Howard Ying, None; Jerry Prince, None
  • Footnotes
    Support None
Investigative Ophthalmology & Visual Science June 2015, Vol.56, 3811. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Andrew Lang, Aaron Carass, Ava K Bittner, Howard S Ying, Jerry L Prince; Automated segmentation of macular SD-OCT scans of retinitis pigmentosa patients shows regional patterns of foveal inner retinal thickening that correlate with visual function. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):3811.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose: To adapt an algorithm previously developed for healthy eyes to segment eight retinal layers in retinitis pigmentosa (RP) OCT data and to correlate the algorithm results with visual function.

Methods: SD-OCT volumes from 17 RP subjects were acquired using a Spectralis scanner (Heidelberg Engineering, Heidelberg, Germany) over a 6x6 mm area of the macula with 512 A-scans and between 19 and 49 B-scans per subject. At the same visit, we assessed ETDRS visual acuity (VA), Pelli-Robson contrast sensitivity (CS), and visual fields (VF) with the Goldmann perimeter. Eight retinal layers were segmented using a previously developed algorithm with the following steps: RPE boundary estimation, intensity normalization, boundary classification, and a graph search algorithm for the final segmentation. Adaptation for RP data required alteration of the intensity normalization, classifier features, and graph constraints. We used a leave-one-out cross-validation scheme to evaluate segmentation performance on nine subjects having manual segmentations.

Results: The average absolute error over all nine boundaries is 4.22 ± 2.44 μm. The photoreceptor boundaries had average errors of less than 4.4 μm. The RNFL-GCL and OPL-ONL interfaces had the largest average errors at 6.25 and 6.43 μm, respectively. Eyes with mixed areas of significantly increased and reduced GCL+IPL thickness near the fovea had statistically significantly smaller Goldmann VFs than eyes with only increased GCL+IPL thickness centrally (P=0.03). There was a statistically significant quadratic relationship between foveal GCL+IPL thickness and Goldmann V4e log retinal area (P=0.006), indicating an initial increase in GCL+IPL thickness, followed by a reduction as VF loss progresses. We found statistically significant relationships between the foveal, extrafoveal or mean macular GCL+IPL thickness and VA (P=0.005-0.009) or CS (P=0.03).

Conclusions: We developed an automated segmentation algorithm which finds eight retinal layers in RP data with good performance, especially in the outer layers which degenerate in RP. It is interesting that inner retinal thickness (GCL+IPL) is also significantly correlated with RP visual function, and regional patterns of foveal inner retinal thickening may represent distinct stages of retinal remodeling that occurs in the GCL+IPL.

×
×

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.

×