Borroni D, Beech M, Williams B, Liu R, Zhao Y, Ma B, Romano V, Alam U, Qi H, Kaye SB, Zheng Y. Building a validated in vivo confocal microscopy (IVCM) dataset for the study of corneal nerves. Investigative Ophthalmology & Visual Science. 2018 Jul 13;59(9):5719-.
https://iovs.arvojournals.org/article.aspx?articleid=2693027
Abstract
Purpose
Corneal nerves are of great importance to both clinicians and scientists. In vivo confocal microscopy (IVCM) enables the non-invasive examination of corneal nerves. This allows the study of nerve alterations in different ocular diseases and in systemic diseases. However, there has been no corresponding increase in the availability of software tools to support the automatic analysis of corneal nerves. The primary reason for this is the lack of a validated dataset to support the development and evaluation of such tools. We propose to build and evaluate a new IVCM dataset for the automatic detection of corneal nerves.
Methods
127 IVCM images were obtained using the Heidelberg Retina Tomograph 3/Rostock Cornea Module (Heidelberg Engineering, Heidelberg, Germany) from healthy normal corneas and corneas with various conditions. The corneal nerves in each image were manually traced independently by two clinical ophthalmologists (DB and MB) who were trained for the annotation task on an in-house Matlab programme (Matlab R14; The Mathworks Inc., Natick, MA). In order to evaluate repeatability, all the images were traced again by the same two observers after a two week period. Intra- and inter-observer agreement were assessed using mean Hausdorff Distance (mHD), which describes, in terms of microns (um), the pointwise minimal distance between two annotations. Since our data contains nerves which are up to in diameter, a mHD value of was deemed as excellent agreement.
Results
Excellent results were achieved for both inter- and intra-observer agreement. The mean mHD distance to observer one’s first annotation from his second was and similarly the mean mHD for the second observer’s annotations was . Between the two graders, the distances were and for the first and second annotations respectively.
Conclusions
Our results have demonstrated that this dataset is valid for the development of automated techniques for the detection of corneal nerves in IVCM images. We are planning to release this dataset to the research community that is interested in the study of corneal nerves in relation to eye disease and systemic disease.
