April 2009
Volume 50, Issue 13
ARVO Annual Meeting Abstract  |   April 2009
Semi-Automated Montaging of the Entire Corneal Sub-Basal Plexus
Author Affiliations & Notes
  • J. Turuwhenua
    Auckland Bioengineering Institute,
    The University of Auckland, Auckland, New Zealand
  • D. V. Patel
    Department of Ophthalmology,
    The University of Auckland, Auckland, New Zealand
  • C. N. J. McGhee
    Department of Ophthalmology,
    The University of Auckland, Auckland, New Zealand
  • Footnotes
    Commercial Relationships  J. Turuwhenua, None; D.V. Patel, None; C.N.J. McGhee, None.
  • Footnotes
    Support  None.
Investigative Ophthalmology & Visual Science April 2009, Vol.50, 3696. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      J. Turuwhenua, D. V. Patel, C. N. J. McGhee; Semi-Automated Montaging of the Entire Corneal Sub-Basal Plexus. Invest. Ophthalmol. Vis. Sci. 2009;50(13):3696.

      Download citation file:

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

  • Supplements

Purpose: : In vivo confocal microscopy (IVCM) has recently enabled elucidation of the 2-dimensional structure of the corneal sub-basal nerve plexus. However, acquiring a full montage of the cornea is a time consuming process: a typical map requires many hundreds of images. The purpose of this work was to develop and test a system for semi-automated montaging of images taken of the living human corneal sub-basal nerve plexus.

Methods: : Software was developed to process a complete set of IVCM images from a normal living human cornea. A preprocessing step removed images containing unwanted features such as epithelium. For the remaining images, keypoints were generated that encoded local branching information. That information was used to generate a potential stitch between pairs of images using the RANSAC algorithm. Image based criteria were applied to decide whether or not the stitches were valid. The resulting adjacency information yielded groupings of overlapping images. Semi-automated means were employed to allow the inclusion of image matches that were below threshold, as well as possible matches with rejected images. The resulting automated montages were compared against two previously manually stitched montages, constructed from a total pool of 640 and 373 images respectively.

Results: : Over 71% of all images were accepted for stitching, with >87% sensitivity and <37% specificity. The low specificity occurred as redundant images had been eliminated from the manually stitched montages. Potentially mismatched images (false positives) were detected (<1.5%), but these were traced to errors in the manually created montages. The adjusted false positive rate was therefore zero in both instances. The final montages both contained 61 groups. Two large groupings were found in each example, containing 68% and 50% of accepted images (367 and 166 images respectively). Unmatched images comprised <11% of all groups. The semi-automated process of adding image pairs manually accounted for <29% reductions in the number of groups. Processing times for these examples were 2 hr 24 min. and 1 hr 7 min., without optimization.

Conclusions: : The majority of accepted images were matched (>89%), half or more were collected into a single group. The semi-automated result agreed with a manually stitched montage. In fact some errors in the latter were revealed by the comparison. Semi-automated stitching will be a useful, effective and time saving tool for further studies involving corneal nerve imaging by IVCM. Further work will aim to extend and optimize the methods developed here.

Keywords: image processing • cornea: basic science 

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.