EARLY KERATOCONUS DISEASE DETECTION USING ORBSCAN II CORNEAL TOPOGRAPHY

Abstract

Keratoconus is an eye disease causing progressive corneal thinning. At an early stage (fruste keratoconus), the symptoms can overlap with those of other eye disorders, making the diagnosis difficult. This led us to propose a new image processing pipeline to automatically calculate the main descriptors used by Ophthalmologists to detect keratoconus, and then classify these data in order to propose an intelligent system able to help specialists in the early recognition of this pathology. To accomplish this, we elaborated a new benchmark database from a corneal topographic Orbscan II device. For keratoconus classification, five different machine learning methods are tested on our new locally collected database, which are: K-Nearest Neighbors (K-NN), Support Vector Machines (SVM), Decision Trees (DT), and two neural networks classifiers (the Radial Basis Function (RBF) and the Multi-Layer Perceptron (MLP)). Experimental results indicate that all classifiers achieved good precision when using all descriptors (the numerical parameters given by the ORBSCAN II topographer and the descriptors obtained after an image processing of the topography map). Furthermore, the SVM outperformed the other classifiers with an accuracy rate of 95.04%. These results were confirmed and validated by a group of experts in ophthalmology and prove the efficiency of our system and the coherence of our new database.