Dictionary Learning-Based Volumetric Image Classification for the Diagnosis of Age-Related Macular Degeneration

A discriminative dictionary-based approach to supporting the classification of 3D Optical Coherence Tomography (OCT) retinal images, so as to determine the presence of Age-related Macular Degeneration (AMD), is described. AMD is one of the leading causes of blindness in people aged over 50 years. The proposed approach is founded on the concept of a uniform 3D image decomposition into a set of sub-volumes where each sub-volume is described in terms of a “spatial gradient” histogram, which in turn is used to define a set of feature vectors (one per sub-volume). Feature selection is conducted using the maximum sum of the squared values of each feature vector for each sub-volume. After that, a “coding-pooling” framework is applied so that each image is represented as a single feature vector. The “coding-pooling” framework generates a representative subset of feature vectors called a dictionary, and then use this dictionary as a guide for the generation of a single feature vectors for each volume. Experiments conducted using the proposed approach, in comparison with range of alternatives, indicated that the approach outperformed other existing methods with an accuracy of 95.2%, sensitivity of 95.7% and specificity of 94.6%.