Detection of keratoconus in anterior segment photographed images using corneal curvature features

Keratoconus is a corneal ectatic disorder with complex aetiology and may induce mild to severe visual impairment and consequently decrease the quality of life. This paper presents a new keratoconus detection method using corneal curvature features to differentiate normal and keratoconus cases. In this study, the eye images known as anterior segmented photographed images (ASPIs) are captured from side view using a smartphone’s camera. For the side-view images, the corneal curvature is segmented using spline function to measure the corneal curvature. A template disc method is implemented to quantitatively measure the steepening of the corneal curvature of the captured ASPIs. Parameters obtained from three different template disc methods, namely, nonlinear, , crossover point, , and trigonometric, , are investigated to represent the most suitable curvature feature. SVM is then employed to classify normal and keratoconus eyes. Results reveal that a standalone nonlinear method gives a reliable parameter with 90% accuracy in classifying the data. However, the classification performance has increased to 99.5% accuracy with the use of all combined features known as a feature vector, . Additionally, classification with the proposed  has successfully distinguished normal and keratoconus cases with sensitivity and specificity rates of 99% and 100%, respectively. The results portray the bright potential of this method in assisting experts during ocular screening specifically to detect keratoconus disease.