Retina Image Vessel Segmentation Using a Hybrid CGLI Level Set Method

Abstract

As a nonintrusive method, the retina imaging provides us with a better way for the diagnosis of ophthalmologic diseases. Extracting the vessel profile automatically from the retina image is an important step in analyzing retina images. A novel hybrid active contour model is proposed to segment the fundus image automatically in this paper. It combines the signed pressure force function introduced by the Selective Binary and Gaussian Filtering Regularized Level Set (SBGFRLS) model with the local intensity property introduced by the Local Binary fitting (LBF) model to overcome the difficulty of the low contrast in segmentation process. It is more robust to the initial condition than the traditional methods and is easily implemented compared to the supervised vessel extraction methods. Proposed segmentation method was evaluated on two public datasets, DRIVE (Digital Retinal Images for Vessel Extraction) and STARE (Structured Analysis of the Retina) (the average accuracy of 0.9390 with 0.7358 sensitivity and 0.9680 specificity on DRIVE datasets and average accuracy of 0.9409 with 0.7449 sensitivity and 0.9690 specificity on STARE datasets). The experimental results show that our method is effective and our method is also robust to some kinds of pathology images compared with the traditional level set methods.

Efficient BFCN for Automatic Retinal Vessel Segmentation

“…The proposed method can automatically configure selectivity in a prototype mode check. Chen [ 13 ] proposed a novel hybrid active contour model for automatic segmentation of fundus images.…”

RFARN: Retinal vessel segmentation based on reverse fusion attention residual network

“…This largely limits the ease of application of these methods in clinical practice. In addition to this, there are methods such as region growth algorithms, EM algorithms with maximum entropy [18], and hybrid active contour models [19] which have been used for the segmentation of fundus images. The unsupervised method described above identifies target vessels by using the intrinsic association between features without training a classifier, and the method is relatively simple to operate requiring less hardware environment for experiments.…”

Recent Techniques and Trends for Retinal Blood Vessel Extraction and Tortuosity Evaluation: A Comprehensive Review

“…The authors use segmentation techniques such as active contours or snake contours. Active contours can be further divided into, for example, Chan Vase, LBF (Local Binary Fitting) and active contours driven by local Gaussian distribution fitting energy, which describes the local intensity of the image with different deviations and diameters [45,49].…”

“…Another method classed as a deformable model is the level set method based on information on local clusters in regions that form a non-homogeneous image of the retina [47]. Chen et al use a combination of a level set function with established selective binary and Gaussian filtering in combination with LBF to work with low contrast images [45].…”