April 2014
Volume 55, Issue 13
Free
ARVO Annual Meeting Abstract  |   April 2014
A fully-automatic fast technique to trace sub-basal layer nerves in corneal images
Author Affiliations & Notes
  • Pedro Guimaraes
    University of Padova, Padova, Italy
  • Jeff Wigdahl
    University of Padova, Padova, Italy
  • Enea Poletti
    University of Padova, Padova, Italy
  • Neil S Lagali
    Linkoping University, Linkoping, Sweden
  • Alfredo Ruggeri
    University of Padova, Padova, Italy
  • Footnotes
    Commercial Relationships Pedro Guimaraes, None; Jeff Wigdahl, None; Enea Poletti, None; Neil Lagali, None; Alfredo Ruggeri, None
  • Footnotes
    Support None
Investigative Ophthalmology & Visual Science April 2014, Vol.55, 2454. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Pedro Guimaraes, Jeff Wigdahl, Enea Poletti, Neil S Lagali, Alfredo Ruggeri; A fully-automatic fast technique to trace sub-basal layer nerves in corneal images. Invest. Ophthalmol. Vis. Sci. 2014;55(13):2454.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose: To develop a robust and fast algorithm capable of tracing the sub-basal plexus nerves from human corneal confocal images.

Methods: A total of 1134 confocal microscopy images covering a field of 400x400 μm (384x384 pixels) of the sub-basal corneal nerve plexus were acquired using the Heidelberg Retina Tomograph (HRT-III) with Rostock Cornea Module (Heidelberg Engineering GmbH, Heidelberg, Germany). All the images were first corrected for uneven illumination and contrast by top-hat filtering. The corrected images were then filtered with a bank of Log-Gabor even and odd kernels with unique orientation-scale pairs. The bank covers a wide enough range of scales and orientations to fully describe the whole nerve network. Each value in the filtered image is defined as the difference between the even and odd maximal filter responses. The filtered image is then segmented by hysteresis thresholding. Finally, fragmented vessel segments are connected by morphological dilation of the resulting binary images, while isolated small segments are erased. All algorithm parameters were optimized on a set of 50 randomly chosen images, while the others were used for testing (N=842). All images were also manually segmented by a cornea specialist, who traced the centerlines of all visible nerves using the NeuronJ tracing plugin for ImageJ, and these results were used as ground-truth reference.

Results: To evaluate the algorithm's performance, the nerve tracings obtained by the proposed automatic technique were compared with the reference ones. On the 842 images of our testing set, a sensitivity of 0.84 ± 0.08 was achieved, with a false detection rate of 0.11 ± 0.07. The time required to analyze a single image was 0.29 ± 0.01 s.

Conclusions: The proposed algorithm proved capable to correctly trace most of the corneal nerves. The achieved quality and processing time appear adequate for the possible application of this technique to clinical practice.

Keywords: 484 cornea: stroma and keratocytes • 550 imaging/image analysis: clinical • 596 microscopy: confocal/tunneling  
×
×

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.

×