Abstract:The automatic extraction of blood vessels in non-fluorescein eye fundus images is a tough task in applications such as diabetic retinopathy screening. However, vessel shapes have complex variations, and accurate modeling of retinal vascular structures is challenging. We have therefore developed a new approach to accurately extract blood vessels in non-fluorescein fundus images using an orientation-aware detector (OAD). The detector was designed according to the intrinsic property of vessels being locally oriented and having linearly elongated structures. We employ the OAD to extract vessel shapes with no assumptions on parametric orientations of vessel shapes. The orientations of vessels can be efficiently modeled by the energy distribution of Fourier transformation. Accordingly, both wide and thin vessels can be extracted with two-scale segmentation in which line operators are applied in large scale and the Gabor filter bank is applied in small scale. A post-processing technique, based on the path opening operation, is applied to eliminate false responses to nonvascular areas, such as retinal structures (optic disc and macula) and pathologies (exudates, hemorrhages,and microaneurysms). This makes the detector robust and structure-aware. By achieving a competitive CAL measurement of 80.82% for the DRIVE database and 68.94% for the STARE, the experimental results demonstrated that the OAD approach outperforms existing segmentation methods. Furthermore, the proposed approach effectively works with non-fluorescein fundus images and proves highly accurate and robust in complicated regions such as the central reflex, close vessels, and crossover points, despite a high level of illumination noise in the original data.

Anisotropic Haralick Edge Detection Scheme with Application to Vessel Segmentation

R-PLUS: a Riemannian Anisotropic Edge Detection Scheme for Vascular Segmentation.

Automatic segmentation of vessels in breast MR sequences as a false positive elimination technique for automatic lesion detection and segmentation using the shape tensor

Segmentation of Retinal Blood Vessel by Merging Shape, Region and Edge Information

A Novel Method for Edge Detection of Retinal Vessels Based on the Model of the Retinal Vascular Network and Mathematical Morphology

2D and 3D Vascular Structures Enhancement via Multiscale Fractional Anisotropy Tensor

A Variational Method for Geometric Regularization of Vascular Segmentation in Medical Images.

Vessel Enhancement Based on Anisotropic Reaction Diffusion

A Shape Induced Anisotropic Flowfor Volumetric Vascular Segmentation in MRA.

Vascular Bifurcation Detection by Symmetric Analysis of Eigenvalues

Intracranial Vascular Segmentation In Phase Contrast Mra Images Using A Shape Driven Level Set Method

Effective statistical edge integration using a flux maximizing scheme for volumetric vascular segmentation in MRA.

Generalization of geometrical flux maximizing flow on Riemannian manifolds for improved volumetric blood vessel segmentation

Curvature Integration in a 5D Kernel for Extracting Vessel Connections in Retinal Images

A Novel Coherence Diffusion Method for Vessel Enhancement

A multistep segmentation algorithm for vessel extraction in medical imaging

Deep matched filtering for retinal vessel segmentation

Edge Morphology Attention Mechanism and Optimal Geometric Matching Connection Model for Vascular Segmentation

Geometric Flows for Vascular Segmentation

An Edge-Based Mesh Segmentation Method for Engineering Objects

Vessel Extraction from Non-Fluorescein Fundus Images Using Orientation-Aware Detector.