Diagnosis and Degree of Evolution in a Keratoconus-Type Corneal Ectasia from Image Processing

Abstract

Keratoconus is a degenerative ocular pathology characterized by the thinning of the cornea, thus affecting many people around the world since this corneal ectasia causes a deformation of the corneal curvature that leads to astigmatism and, in more severe cases, to blindness. Treating physicians use non-invasive instruments, such is the case of Pentacam®, which takes images of the cornea, both the topography and the profile of the cornea, which allows them to diagnose, evaluate and treat this disease; this is known as morphological characterization of the cornea. On the other hand, Berlin/Ambrosio analysis helps in the identification and subsequent diagnosis since this analysis uses a mathematical model of linear progression, which identifies the different curves with the severity of the disease. Therefore, the aim of this study is to use the images provided by Pentacam®, Berlin/Ambrosio analysis, and vision parameters in a convolutional neural network to evaluate if this disparity could be used to help with the diagnosis of keratoconus and, consequently, generate a more precise and optimal method in the diagnosis of keratoconus. As a result, the processing and comparison of the images and the parameters allowed a 10% increase in the results of specificity and sensitivity of the mean and severe stages when combining tools (corneal profile and vision parameters) in the CNN reaching ranges of 90 to 95%. Furthermore, it is important to highlight that in the early-stage study, its improvement was around 20% in specificity, sensitivity, and accuracy.