An Active Contour Model for Segmenting and Measuring Retinal Vessels

Abstract

This paper presents an algorithm for segmenting and measuring retinal vessels, by growing a "Ribbon of Twins" active contour model, which uses two pairs of contours to capture each vessel edge, while maintaining width consistency. The algorithm is initialized using a generalized morphological order filter to identify approximate vessels centerlines. Once the vessel segments are identified the network topology is determined using an implicit neural cost function to resolve junction configurations. The algorithm is robust, and can accurately locate vessel edges under difficult conditions, including noisy blurred edges, closely parallel vessels, light reflex phenomena, and very fine vessels. It yields precise vessel width measurements, with subpixel average width errors. We compare the algorithm with several benchmarks from the literature, demonstrating higher segmentation sensitivity and more accurate width measurement.

Automated Measurements of Retinal Bifurcations

“…The sources of biological scatter include the variable curvature, and non-uniform width of vessels [4]. The automated algorithm calculates the vessel width integrated along a short length of the vessel at the bifurcation [10], and has been shown in evaluation [14] to provide more accurate width measurement than alternative algorithms.…”

“…2. The algorithm is robust, and can accurately locate vessel edges under difficult conditions and yields precise vessel width measurements, with subpixel average width errors [10].…”

A simplified fluid–structure model for arterial flow. Application to retinal hemodynamics

“…The original image was taken in the Drive dataset [21]. The vasculature has been segmented by applying the imaging methods presented in [22,23]. The 2D data have then been expanded into a 3D-network by assuming a circular section and projecting the results onto a sphere representing the eye.…”

Improved automated vessel segmentation for diagnosing eye diseases using fundus images

“…Examples of such approaches include machine learning, [3][4][5][6][7][8][9][10][11][12] matched filtering, [13,14] vessel tracking, [15,16] mathematical morphology, [17] model approaches, [18,19] and connected operators. [20,21] Machine-learning methods assign one or more groups to pixels in the retinal image, using multiple numeric pixel features to group them.…”