Automated macular pathology diagnosis in retinal OCT images using multi-scale spatial pyramid with local binary patterns

Abstract

We address a novel problem domain in the analysis of optical coherence tomography (OCT) images: the diagnosis of multiple macular pathologies in retinal OCT images. The goal is to identify the presence of normal macula and each of three types of macular pathologies, namely, macular hole, macular edema, and age-related macular degeneration, in the OCT slice centered at the fovea. We use a machine learning approach based on global image descriptors formed from a multi-scale spatial pyramid. Our local descriptors are dimension-reduced Local Binary Pattern histograms, which are capable of encoding texture information from OCT images of the retina. Our representation operates at multiple spatial scales and granularities, leading to robust performance. We use 2-class Support Vector Machine classifiers to identify the presence of normal macula and each of the three pathologies. We conducted extensive experiments on a large dataset consisting of 326 OCT scans from 136 patients. The results show that the proposed method is very effective.

Fig. 1

(a) Example 3D macular OCT scan. (b)(c)(d) Example x-z slice with (b) MH (red), ME (blue), and AMD (green), (c) shadowing effects, (d) a detached tissue.

Fig. 2

Characteristics of (a) normal macula: a smooth depression shows at the center (fovea), (b) macular hole (MH): a partial or full hole formation at the fovea, (c) macular edema (ME) : retinal thickening and liquid accumulation appears as black blobs around the fovea, (d) age-related macular degeneration (AMD): irregular contours usually extruded in dome shapes appear at the bottom layer of the retina.

Fig. 3

Stages in the constructing of the global image descriptor.

Fig. 4

Structures of a 3-level MSSP and SP. The green lines indicate the overlapped blocks which can also be added in the global representation.

Fig. 5

Examples of (a) LBP_8,1 code computation, (b) uniform and non-uniform LBP.

Fig. 6

Visualization of ${\mathit{LBP}}_{8,1}^{u2}$ and ${\mathit{LBP}}_{8,1}^{\neg u2}$ on example level-0 images. All ${\mathit{LBP}}_{8,1}^{\neg u2}$ are shown in red in the 2nd row, and the individual ${\mathit{LBP}}_{8,1}^{\neg u2}$ codes are shown in gray-level in the 3rd row. We find that although ${\mathit{LBP}}_{8,1}^{\neg u2}$ patterns have low counts, most of them reside in the important contours and can therefore be useful for discrimination.