RNN-PSO: a tuned neural network with optimisation algorithm for keratoconus classification

Keratoconus is a medical illness in which the cornea attains a conical shape due to the thinning of the corneal stroma. Its symptoms vary based on the keratoconus stage with early phases going unnoticed, while the advanced phases are characterised by vision loss and protrusion. The diagnosis of keratoconus and its severity classification using corneal topography images has gained importance with advancements in imaging technology and machine learning. Convolutional Neural Networks (CNN) is used for extracting significant features from the images to categorise the keratoconus. Then, the extracted features are classified by using a Recurrent Neural Network (RNN), and the hyperparameters of the RNN are optimised by using Particle Swarm Optimisation (PSO) which improves the keratoconus classification performance. The performance of the proposed method is evaluated with four classes namely, normal, sub-clinical, keratoconus and advanced keratoconus in terms of accuracy, sensitivity and specificity. The method attains an accuracy of 0.80 on normal, 0.81 on sub-clinical, 0.85 on keratoconus and 0.91 on advanced keratoconus.