[Paper Review] Automated Segmentation of Retinal Layers from Optical Coherent Tomography Images Using Geodesic Distance

Jinming Duan, Christopher R. Tench+3 more|arXiv (Cornell University)|Sep 7, 2016

Glaucoma and retinal disorders25 references3 citations

TL;DR

This paper proposes a novel geodesic distance method (GDM) for automated segmentation of retinal layers in 2D and 3D optical coherence tomography (OCT) images. By integrating horizontal and vertical intensity gradients through an exponential weighting function and solving the Eikonal equation via fast sweeping, GDM achieves robust segmentation of complex, curved, and pathological retinal structures with higher accuracy than active contour and graph-based methods.

ABSTRACT

Optical coherence tomography (OCT) is a non-invasive imaging technique that can produce images of the eye at the microscopic level. OCT image segmentation to localise retinal layer boundaries is a fundamental procedure for diagnosing and monitoring the progression of retinal and optical nerve disorders. In this paper, we introduce a novel and accurate geodesic distance method (GDM) for OCT segmentation of both healthy and pathological images in either two- or three-dimensional spaces. The method uses a weighted geodesic distance by an exponential function, taking into account both horizontal and vertical intensity variations. The weighted geodesic distance is efficiently calculated from an Eikonal equation via the fast sweeping method. The segmentation is then realised by solving an ordinary differential equation with the geodesic distance. The results of the GDM are compared with manually segmented retinal layer boundaries/surfaces. Extensive experiments demonstrate that the proposed GDM is robust to complex retinal structures with large curvatures and irregularities and it outperforms the parametric active contour algorithm as well as the graph theoretic based approaches for delineating the retinal layers in both healthy and pathological images.

Motivation & Objective

• To address the challenge of accurately segmenting retinal layer boundaries in OCT images affected by speckle noise, intensity inhomogeneity, and pathological irregularities.
• To improve segmentation robustness in regions with large curvatures and weak or low-contrast boundaries, especially in the foveal depression.
• To develop a computationally efficient and accurate method for both 2D B-scan and 3D volumetric OCT image segmentation.
• To overcome limitations of existing methods such as local minima in active contours and high computational cost in graph-based approaches.

Proposed method

• Uses a weighted geodesic distance computed via an Eikonal equation, with an exponential function that integrates both horizontal and vertical intensity gradients.
• Employs the fast sweeping method to efficiently solve the Eikonal equation for 3D distance computation, enabling fast and stable numerical solution.
• Applies a time-dependent gradient descent equation to detect retinal layer boundaries based on the computed geodesic distance.
• Implements a local search region around the detected boundary to refine and delineate all nine retinal layer boundaries, mitigating local minima issues.
• Utilizes a Godunov upwind finite difference scheme for discretization of the Eikonal equation with proper boundary condition handling.
• Applies alternating sweeping orders in a fast Gauss-Seidel iteration to converge to the unique solution of the Eikonal equation.

Experimental results

Research questions

• RQ1 Can a geodesic distance method that incorporates both horizontal and vertical intensity gradients improve segmentation accuracy in OCT images with complex retinal structures?
• RQ2 How does the proposed GDM perform in segmenting retinal layers in pathological OCT images with large curvatures and low-contrast boundaries?
• RQ3 Does the exponential weighting function enhance foveal depression regions and weak boundaries more effectively than single-direction gradient methods?
• RQ4 How does the GDM compare in accuracy and computational efficiency to state-of-the-art methods such as parametric active contours and graph-theoretic approaches?
• RQ5 Can the local search refinement strategy effectively overcome local minima and improve boundary delineation in challenging regions?

Key findings

• The proposed GDM achieves higher segmentation accuracy than the parametric active contour model and graph-theoretic approaches on both healthy and pathological OCT images.
• The method demonstrates robustness in segmenting retinal layers with large curvatures and irregularities, particularly in the foveal region where boundaries are subtle.
• The exponential weighting function effectively enhances weak and low-contrast boundaries, improving detection in regions with poor contrast or shadowing.
• The fast sweeping method enables efficient solution of the Eikonal equation, making the algorithm suitable for 3D volumetric OCT data processing.
• The local search refinement step successfully mitigates local minima issues, leading to more consistent and accurate boundary detection across all nine retinal layers.
• Although not directly comparable due to implementation differences, the GDM outperforms Dufour’s graph method in accuracy, despite Dufour’s method requiring 14.68 seconds on a 496×633×10 volume.