ABSTRACT
In this book chapter, we present a computer-aided diagnostic (CAD) system for early detection of diabetic retinopathy (DR) from optical coherence tomography (OCT) images. The proposed CAD system consists of three main steps. In the first step, 12 distinct retinal layers are localized and segmented based on a novel joint model that combines shape, intensity, and spatial information. The shape prior is built using a subset of co-aligned training OCT images. The shape model is then adapted during the segmentation process using visual appearance characteristics that are described using both pixel-wise image intensities and their spatial interaction features. The intensities are modeled using a linear combination of discrete Gaussians (LCDG) model, whereas the spatial information is modeled using a second-order Markov Gibbs random field (MGRF) model that accounts for noise and inhomogeneities. In the second step, three features, namely the reflectivity, curvature, and thickness of the retinal layers, are derived from the segmented OCT images, which can be used to quantitatively distinguish normal and diabetic subjects. Then the cumulative distribution function (CDF) values of the locally extracted features for each segmented layer are constructed. Finally, a multistage Deep Fusion Classification Network (DFCN), trained by stacked layers of Non-Negativity Constraint Autoencoder (NCAE), is used to classify the subject as normal or diabetic based on the CDFs for the most discriminant layers of the retina. Preliminary experiments on 52 clinical OCT scans resulted in 100% correct classification, indicating the high accuracy of the proposed framework and holding promise of the proposed CAD system as a reliable non-invasive diagnostic tool.
Keywords: OCT, Diabetic Retinopathy, shape prior, CAD, LCDG, MGRF.
