Classification of corneal layers in confocal microscopy

A confocal microscope can produce grayscale images of the different layers of the cornea. The images must then be classified, i.e. the layer recognized, using the shape of the cells contained, which is uniquely related to each specific layer. A method was developed, in which the image is at first divided into portions and a separate classification is carried out for each portion. It is based first on the binarization of the image (a prerequisite for the successful application of the following step) and then on the description of the cell shape by means of Hu variables (central moments). An artificial neural network (ANN) is then used to classify each portion according to the values assumed by these variables. The responses for all the image portions are eventually combined to derive the classification of the whole image. A Matlab prototype of the classification system was developed considering images of three corneal layers (Bowman, stroma, endothelium) in normal subjects. The system was tested on a preliminary validation set of 46 images and good results were obtained. To avoid critical step of binarization, an alternative for cell shape description was investigated, based on Zernike moment invariants. A new ANN was developed and trained. The results achieved were better than the ones obtained with the previous technique and no binarization was necessary.