Neuro-fuzzy classifier for corneal nerve images

Abstract

Alterations in the corneal nerves have been associated with changes in the neuropathy condition of a patient suffering from chronic diseases. A corneal confocal microscope provides a non-invasive way to capture a series of images from the corneal sub-basal nerve plexus. These images undergo a tedious process of manual analysis before the classification of the state of nerves is determined as normal or abnormal. To address this limitation, we introduce a pioneering technique for automating corneal nerve image classification using Adaptive Neuro-Fuzzy Inference System. Prior to image classification, the images are preprocessed using discrete wavelet transform, filtering and morphological operations. The resulting segmented image is used to produce a feature set representative of the image. This is followed by training the neuro fuzzy classifier on the extracted features. The trained classifier is then used for predicting the state of the nerves in the images. Initial experiments yield a classification accuracy of 0.86 reflecting the effectiveness of the proposed technique.