Two-dimensional segmentation of the retinal vascular network from optical coherence tomography

Abstract

Abstract. The automatic segmentation of the retinal vascular network from ocular fundus images has been performed by several research groups. Although different approaches have been proposed for traditional imaging modalities, only a few have addressed this problem for optical coherence tomography (OCT). Furthermore, these approaches were focused on the optic nerve head region. Compared to color fundus photography and fluorescein angiography, two-dimensional ocular fundus reference images computed from three-dimensional OCT data present additional problems related to system lateral resolution, image contrast, and noise. Specifically, the combination of system lateral resolution and vessel diameter in the macular region renders the process particularly complex, which might partly explain the focus on the optic disc region. In this report, we describe a set of features computed from standard OCT data of the human macula that are used by a supervised-learning process (support vector machines) to automatically segment the vascular network. For a set of macular OCT scans of healthy subjects and diabetic patients, the proposed method achieves 98% accuracy, 99% specificity, and 83% sensitivity. This method was also tested on OCT data of the optic nerve head region achieving similar results.

Three-dimensional segmentation and reconstruction of the retinal vasculature from spectral-domain optical coherence tomography

“…To obtain the 2-D vascular network segmentation we resort to an approach previously publish by Rodrigues et al 23 For clarity, we shortly describe in this section the used algorithm and the obtained results.…”

“…This reference is then used to compute features that are able to discriminate each pixel into the vessel or non-vessel groups. 23 The A-scans whose pixels were classified into the vessel group, i.e., A-scans that intersect blood vessels, are then processed to determine the depth-wise location of the vessel. Fig.…”

“…Specifically, intensity-based features, local-phase features, and features computed with Gaussian-derivative filters, log-Gabor and average filter banks, and band-pass filters. 23 A supervised learning algorithm (support vector machine) was then used to train and classify each A-scan of the OCT volume V into vessel or non-vessel A-scans, i.e., to discriminate between A-scans that cross retinal blood vessels and the remaining A-scans. Formally, we define the set A = {A i (.)…”

Retinal Biomarkers of Alzheimer’s Disease: Insights from Transgenic Mouse Models

“…These approaches share the idea of using depth-wise averaging (total or partial) of each individual Ascan [20]. In this work, for each of the three layers (Layer 1, Layer 2 and Layer 3) a mean value fundus (MVF) image [21] is computed by (2) as the average of the A-scan values within the respective layer.…”

“…where V is the OCT volume (of size 512×512×1024 voxels), Z 1 i (x, y) and Z 2 i (x, y) are the limits of Layer i (i = {1, 2, 3}) at coordinates (x, y) and the coordinate system for the OCT data is defined as: x is the nasal-temporal direction, y is the superior-inferior direction, and z is the anterior-posterior (depth) direction [21].…”