Investigative Ophthalmology & Visual Science Cover Image for Volume 58, Issue 8
June 2017
Volume 58, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2017
Automated analysis of corneal ulcer from external photography of the eye
Author Affiliations & Notes
  • Tapan Patel
    Ophthalmology, Kellogg Eye Center, Ann Arbor, Michigan, United States
  • N.Venkatesh Prajna
    Aravind Eye Care System, Madurai, India
  • Lakshey Dudeja
    Aravind Eye Care System, Madurai, India
  • Nita Valikodath
    Ophthalmology, Kellogg Eye Center, Ann Arbor, Michigan, United States
  • Maria A Woodward
    Ophthalmology, Kellogg Eye Center, Ann Arbor, Michigan, United States
  • Footnotes
    Commercial Relationships   Tapan Patel, None; N.Venkatesh Prajna, None; Lakshey Dudeja, None; Nita Valikodath, None; Maria Woodward, None
  • Footnotes
    Support  NIH K23K23EY023596
Investigative Ophthalmology & Visual Science June 2017, Vol.58, 3528. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Tapan Patel, N.Venkatesh Prajna, Lakshey Dudeja, Nita Valikodath, Maria A Woodward; Automated analysis of corneal ulcer from external photography of the eye. Invest. Ophthalmol. Vis. Sci. 2017;58(8):3528.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Accurate assessment of corneal ulcer size is essential for monitoring response to therapy. This relies on the ophthalmologist’s estimation of the size of the epithelial defect (ED) and stromal infiltrate (SI) at the slit-lamp, and is augmented by acquiring external photographs of the ulcer. We introduce an automated method to construct quantitative models of the corneal ulcer from these images, which will facilitate more consistent therapeutic monitoring.

Methods : Fourteen patients with a corneal ulcer were examined by 4 separate ophthalmologists at the slit lamp and an expert consensus of the size of the ED and SI were recorded. An external photo of the affected eye was also captured.

We developed a graphical user interface to facilitate semi-automated segmentation (delineation) of ED and SI from external photography (Fig 1A). The user initializes seed regions in the foreground (ED or SI) and the surrounding background regions (Fig 1B; foreground in blue, background in red). Random forest tissue classification is used to generate a probability map for the foreground image (Fig 1C). A level-set segmentation is subsequently performed, in which the probability map generated by random forest is used to drive active contour evolution to delineate the boundary of the ED or SI (Fig 1D-F). The area of the ED or SI is then automatically computed.

Separately, each photograph was manually traced and the Dice similarity coefficient between manual and automated models computed to assess the accuracy of automated segmentation. We also calculated the intra-class correlation coefficient (ICC) to compare the agreement in ED and SI size between automatically measured and expert consensus.

Results : The Dice similarity coefficient between manual and automated segmentation of ED and SI were 0.92 ± 0.03 and 0.89 ± 0.05, respectively (N=14, mean ± SD). The ICC comparing agreement in ED and SI size between automated and expert consensus were 0.96 (95% CI [0.89, 0.98]) and 0.82 (95% CI [0.55, 0.93]), respectively.

Conclusions : We describe a novel method for semi-automated analysis of corneal ulcers. Automated measurements of ED and SI size are in good agreement with expert consensus and have the potential to enhance the reliability of therapeutic monitoring.

This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.

 

Figure 1: A: external photograph of corneal ulcer. B: seed regions. C: probability map. D-E: active contour evolution. F: final segmentation.

Figure 1: A: external photograph of corneal ulcer. B: seed regions. C: probability map. D-E: active contour evolution. F: final segmentation.

×
×

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.

×