Artificial Intelligence Hybrid System for Enhancing Retinal Diseases Classification Using Automated Deep Features Extracted from OCT Images
Amin Alqudah, Ali Mohammad Alqudah, Ma'moun AlTantawi
Abstract
After the advent of eye 2D imaging technology, Optical Coherence Tomography (OCT) has become one of the most effective and commonly used imaging techniques for non-invasive retinal eye disease evaluation. Blindness is primarily diagnosed using OCT with one of the following two eye diseases categories: diabetic macular edema (DME) or age-related macular degeneration (AMD). The classification of eye retina diseases using OCT images has recently become a challenge with the development of machine learning and profound learning techniques. In this paper, a hybrid artificial intelligence system for multiclass classification of eye retina diseases has been proposed, using automated deep features extracted, based on Advanced OCT Network (AOCTNet) architecture from OCT images especially spectral-domain (SD-OCT) images. The proposed methodology mainly can be used to classify retinal diseases into normal and four abnormal classes (AMD, choroidal neovascularization (CNV), DME, and Drusen). The proposed system was constructed using eight types of machine learning algorithms (Support Vector Machine with Linear kernel (LSVM), Support Vector Machine with Radial Basis Function kernel (RBF SVM), Artificial Neural Network (ANN), K Nearest Neighbor (KNN), Random Forest (RF), Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), and Nave Bayes (NB)). The classifiers achieved high performance. For example, the KNN and the RF classifiers achieved an accuracy of 99.44 and 99.12, respectively. This methodology is potentially a powerful computer-aided diagnostic (CAD) tool for the use of SD-OCT imaging for retinal diseases.