The corneal endothelium serves as the posterior barrier of the cornea. Factors such as clarity and refractive properties of
the cornea are in direct relationship to the quality of the endothelium. The endothelial cell density is considered the most
important morphological factor of the corneal endothelium. Pathological conditions and physical trauma may threaten
the endothelial cell density to such an extent that the optical property of the cornea and thus clear eyesight is threatened.
Diagnosis of the corneal endothelium through morphometry is an important part of several clinical applications.
Morphometry of the corneal endothelium is presently carried out by semi automated analysis of pictures captured by a
Clinical Specular Microscope (CSM). Because of the occasional need of operator involvement, this process can be
tedious, having a negative impact on sampling size. This study was dedicated to the development and use of fully
automated analysis of a very large range of images of the corneal endothelium, captured by CSM, using Fourier
analysis. Software was developed in the mathematical programming language Matlab. Pictures of the corneal
endothelium, captured by CSM, were read into the analysis software. The software automatically performed digital
enhancement of the images, normalizing lights and contrasts. The digitally enhanced images of the corneal endothelium
were Fourier transformed, using the fast Fourier transform (FFT) and stored as new images. Tools were developed and
applied for identification and analysis of relevant characteristics of the Fourier transformed images. The data obtained
from each Fourier transformed image was used to calculate the mean cell density of its corresponding corneal
endothelium. The calculation was based on well known diffraction theory. Results in form of estimated cell density of
the corneal endothelium were obtained, using fully automated analysis software on 292 images captured by CSM. The
cell density obtained by the fully automated analysis was compared to the cell density obtained from classical, semi-automated
analysis and a relatively large correlation was found.
|