Con-Ker: A Convolutional Neural Network Based Approach for Keratoconus Detection and Classification

The paper is on the detection of keratoconus a corneal progressive disorder leading to the thinning and also protrusion of the cornea associated with symptoms like astigmatism, increased sensitivity to bright light, glare, clouded vision, eye irritation, and others, In recent times there has been increasing in a number of keratoconus cases. Keratoconus is normally described as a non-inflammatory pathology. The main contribution of the paper is to facilitate detection and also classification of the keratoconus based on the progression using Convolution neural networks. The paper is about the implementation of different CNN algorithms which will classify the disorder based on the progression into 4 different classes. The CNN algorithms analyze the corneal topography of the eye and classify based on the severity of the disorder. We introduce an effective CNN model called CON-KER for the detection and classification of the disorder. Further CNN algorithms like Alexnet and Vgg 19 were implemented for the same. The results show that the CON-KER model has yielded an accuracy of 96.26% compared to other algorithms like vgg19 which yielded 94.76% and AlexNet with 86% accuracy. This work can help by assisting the ophthalmologist in reducing diagnostic errors and also help in the rapid screening of the patients.