September 2016
Volume 57, Issue 12
Open Access
ARVO Annual Meeting Abstract  |   September 2016
Merging Cell Fragments in Oversegmented Corneal Endothelium Images
Author Affiliations & Notes
  • Juan Pedro Vigueras-Guillén
    Rotterdam Ophthalmic Institute, Rotterdam, Netherlands
    Quantitative Imaging Group, Delft University of Technology, Delft, Netherlands
  • Jeroen van Rooij
    Rotterdam Eye Hospital, Rotterdam, Netherlands
  • Angela Engel
    Rotterdam Ophthalmic Institute, Rotterdam, Netherlands
  • Koenraad Arndt Vermeer
    Rotterdam Ophthalmic Institute, Rotterdam, Netherlands
  • Lucas J. van Vliet
    Quantitative Imaging Group, Delft University of Technology, Delft, Netherlands
  • Footnotes
    Commercial Relationships   Juan Vigueras-Guillén, None; Jeroen van Rooij, None; Angela Engel, None; Koenraad Vermeer, None; Lucas J. van Vliet, None
  • Footnotes
    Support  ZonMw 842005002. ZonMw 842005004.
Investigative Ophthalmology & Visual Science September 2016, Vol.57, 5931. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Juan Pedro Vigueras-Guillén, Jeroen van Rooij, Angela Engel, Koenraad Arndt Vermeer, Lucas J. van Vliet; Merging Cell Fragments in Oversegmented Corneal Endothelium Images. Invest. Ophthalmol. Vis. Sci. 2016;57(12):5931.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Biomarkers to assess the corneal endothelium, such as cell density, polymegathism and pleomorphism, are all derived from a segmentation of the endothelial cells. The large variation in cell size makes such segmentation difficult, causing either over- or undersegmentation. We evaluate a framework to combine cell fragments produced by oversegmentation of endothelial cells.

Methods : Five specular microscopy images (Topcon SP-1P) were captured six months post-op from five patients (ages 57-68) who had DSAEK (Descemet Stripping Automated Endothelial Keratoplasty) surgery in 2014-2015. The endothelium images were annotated by an expert to create the ground truth.
A stochastic watershed method (Selig et al., BMC Medical Imaging 15:13, 2015) was employed to generate superpixels, initializing the algorithm with a cell density of 6.000 cells/mm2 to create an oversegmented image.
For each superpixel and for each combination of two adjacent superpixels, area (size) and circularity (shape) features were extracted. By using such features in a 2D Gaussian multivariate model, the probability of being a cell was inferred for each case. If two combined superpixels had a higher probability than both independent superpixels, the merge was established (fig. 1).

Results : The results from the merging algorithm were visually compared with the ground truth. The number of over- and undersegmented cells before and after the merging process were counted manually. The results are summarized in Table 1.
In total, the number of oversegmented cells were reduced by a total of 51.1 %, and 98.8 % of merges were correct. Thus, barely any undersegmented cell was created.

Conclusions : The pair-wise merging technique can reduce the number of oversegmented cells significantly by merely using two features. Considering multiple fragments simultaneously and including additional features might further reduce the amount of oversegmentation.

This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.

 

Table 1. Results on the oversegmented corneal endothelium images in number of under- and oversegmented cells, and cell density estimation error with respect to the ground truth.

Table 1. Results on the oversegmented corneal endothelium images in number of under- and oversegmented cells, and cell density estimation error with respect to the ground truth.

 

Figure 1. Representative example of the merging technique: Among all the combinations of A, A+E has a higher probability of being a cell (based on its area and circularity) than both A and E independently. No other combination generates a better probability. Thus, A and E are merged.

Figure 1. Representative example of the merging technique: Among all the combinations of A, A+E has a higher probability of being a cell (based on its area and circularity) than both A and E independently. No other combination generates a better probability. Thus, A and E are merged.

×
×

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.

×