Corneal images acquired by in-vivo specular microscopy provide clinical information on the cornea endothelium health state. At present, the analysis is based on manual or semi-automatic methods and the segmentation of a large number of endothelial cells is required for a meaningful estimation of the clinical parameters (cell density, pleomorphism, polymegethism). For the practical application in clinical settings, a computerized method capable to fully automatize the segmentation procedure would be needed.
We propose here a supervised classification scheme for the segmentation of endothelium cells. In order to detect the cell contour polygon, i.e., its three components as vertexes, sides and body, a multi-scale 2-dimensional matched filter approach is employed. Three kernels have been specifically designed to the detect the three cell components’
, which are then used as features to train a Support Vector Machine classifier and to provide the final segmentation of the cells.
Performance of the proposed method is assessed by computing on a set of 20 images the differences in the three clinical parameters estimated from automatic segmentation and from manual segmentation. The results confirm that the automated system is capable to provide reliable estimates of these important clinical parameters.