Automatic keratoconus detection by means of a neural network: comparison between a monocular and a binocular approach

The results of the utilisation of neural networks for the automatic identification of keratoconus from corneal maps, using 9 previously selected objective parameters, were compared. The keratoconus is an asymmetrical pathology, sometimes monolateral, while the maps of normal eyes and of congenital astigmatism are more often symmetrical. So we compared two methods, the first (monocular) in which each eye is considered alone and the second (binocular) in which information from both eyes of the same subject are the input of the neural network. The binocular information could in fact improve the ability of the neural network to identify the keratoconus corneotopographic patterns. The comparison of the results obtained with the two approaches does not show a significant difference: a sensitivity of 92% in both cases and a specificity of 99% and 100% respectively were obtained. Therefore the two methods must be considered equivalent.