U-Net Driven Deep Learning Segmentation for Retinal Vasculature for Early Disease Detection

A comprehensive survey on segmentation techniques for retinal vessel segmentation

A novel automated method for the segmentation of blood vessels in retinal images based upon enhancement and maximum entropy thresholding is proposed. Blood vessels usually have poor local contrast. Before thresholding fundus images, several matched filters are employed to enhance the contrast of blood vessels. The matched-filter-response (MFR) image is processed by a thresholding scheme in order to extract blood vessels from the background. Then, the proposed thresholding approach evaluates two-dimensional entropies based on the gray level-gradient cooccurrence matrix. The 2D threshold vector that maximizes the edge class entropies is selected. This thresholding method utilizes the information of gray level and gradient in the MFR image. It is found that the proposed algorithm works well in normal or abnormal retinal images.

In this paper, we present an automatic method based on context-dependent characteristics for the detection and classification of arterial vessels and venous vessels in retinal fundus images. It provides a non-invasive opportunity and effective foundation for the diagnosis of several medical pathologies. In the proposed method, a combination of shifted filter responses is used, which can selectively respond to vessels. It achieves orientation selectivity by computing the weighted geometric mean of the output of a pool of Difference-of-Gaussian filters, whose supports are aligned in a collinear manner. We then configure two combinations of shifted filters, namely symmetric and asymmetric, that are selective for bars and bar-endings, respectively. We achieve vessel detection by summing up the responses of the two filters. Then we extract the morphology and topological characteristics based on the vessel segmentation, and specifically present context-dependent features of blood vessels, including the shape, structure, relative position, context information and other important features. Based on these features, we use JointBoost classifier to construct potential function for conditional random fields (CRFs) model, and train the labeled samples to classify arteriovenous blood vessels in retinal images. The training and testing data sets were prepared according to the results based on DRIVE dataset provided by Estrada et al. The experimental results show that the accuracy of the proposed method for vein and artery detection is 91.1% and 94.5%, respectively, which is superior to that of the state-of-the-art methods. It can be used as a clinical reference for computer-assisted quantitative analysis of fundus images.

Assessment of blood vessels in retinal images is an important factor for many medical disorders. The changes in the retinal vessels due to the pathologies can be easily identified by segmenting the retinal vessels. Segmentation of retinal vessels is done to identify the early diagnosis of the disease like glaucoma, diabetic retinopathy, macular degeneration, hypertensive retinopathy and arteriosclerosis. In this paper, we propose two automatic blood vessel segmentation methods. The first proposed algorithm starts with the extraction of blood vessel centerline pixels. The final segmentation is obtained using an iterative region growing method that merges the contents of several binary images resulting from vessel width dependent modified morphological filters on normalized retinal images. In the second proposed algorithm the blood vessel is segmented using normalized modified morphological operations and neuro fuzzy classifier. Normalized morphological operations are used to enhance the vessels and neuro fuzzy classifier is used to segment retinal blood vessels. These methods are applied on the publicly available DRIVE database and the experimental results obtained by using green channel images have been presented and their results are compared with recently published methods. The results demonstrate that our algorithms are very effective methods to detect retinal blood vessels. DOI: http://dx.doi.org/10.11591/ijece.v4i3.5829

This paper presents an efficient method for automatic segmentation of blood vessels in retinal images. Specifically, we also delineate vascular intersections/crossovers. The proposed algorithm is composed of three steps: matched filter, fuzzy k-median (FKMED), and length filter. The segmentation results are compared with clinically generated vessel segmentation and are evaluated in terms of sensitivity and specificity. The results are encouraging and will be used for further application such as personal identification.

Retinal (eye fundus) images are widely used for diagnostic purposes by ophthalmologists. The normal features of eye fundus images include the optic nerve disc, fovea and blood vessels. Algorithms for identifying blood vessels in the eye fundus image generally fall into two classes: extraction of vessel information and segmentation of vessel pixels. Algorithms of the first group start on known vessel point and trace the vasculature structure in the image. Algorithms of the second group perform a binary classification (vessel or non-vessel, i.e. background) in accordance of some threshold. We focus here on the binarization [4] methods that adapt the threshold value on each pixel to the global/local image characteristics. Global binarization methods [5] try to find a single threshold value for the whole image. Local binarization methods [3] compute thresholds individually for each pixel using information from the local neighborhood of the pixel. In this paper, we modify and improve the Sauvola local binarization method [3] by extending its abilities to be applied for eye fundus pictures analysis. This method has been adopted for automatic detection of blood vessels in retinal images. We suggest automatic parameter selection for Sauvola method. Our modification allows determine/extract the blood vessels almost independently of the brightness of the picture.

This paper describes a methodology for the segmentation of blood vessels in digital images of human eye retina. The proposed method is based on the background subtraction between a filtered retinal image by anisotropic diffusion and an approximation of the retinal background, obtained by a median filtering. The subtraction operation results in an image of differences, which is enhanced by a local histogram stretching and thresholded to detect the blood vessels. Finally, the obtained binary image is filtered aiming to remove small signals and false responses which are related to retina pathologies. We evaluated the proposed method using STARE and DRIVE image sets, in which the results have shown higher accuracy rates when compared with similar approaches.

The high importance of the accurate and early diagnostic has motivated the development of computer vision techniques of image processing and segmentation required for an completely automated assessment system for the retinal conditions. In this study we present a new algorithm built on wavelets transforms and mathematical morphology for detecting the optic disc and we explore the tubular characteristic of the blood vessels to segment the retinal veins and arteries. Both, optic disc and vascular structure, are landmarks for image registration and are essential for the retinal image analysis. Instead of a manual try and error method to choose the best parameters for detecting vessels as accurately as possible, we used a genetic algorithm and its sequence of generations and crossovers. However the technique of exploring the tubular characteristic of the vessels reaches its limits when the vessels are represented by, sometimes not continuous, winding lines of 1 pixel. To overcome this limitation we adopted a graph based approach using Dijkstra's shortest path algorithm to track the segments and a statistic method of Student t distribution to assess whether or not the identified segment is part of the vascular structure. The proposed method was developed and tested on the Digital Retinal Images for Vessel Extraction (DRIVE) freely available database, which contains 40 annotated color eye fundus image.

In this paper, a system for automated analysis of retinal images is proposed. This system segments blood vessels in retinal images and recognizes the main features of the fundus on digital color images. The recognized features were defined as blood vessels, optic disc, and fovea. An algorithm called 2D matched filters response has been proposed for the detection of blood vessels. Also, automatic recognition and localization methods for optic disc and fovea have been introduced and discussed. Moreover, a method for detecting left and right retinal fundus images has been presented.KeywordsOptic DiscRetinal ImageRetinal VesselCentral Retinal ArteryFundus ImageThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Aiming to identify small blood vessels and low-contrast areas in retinal images, the paper presents an innovative approach to segmenting blood vessels in retinal images by employing a multi-scale dual-path convolutional neural network. It utilises Gabor filters to capture the unique characteristics of vessels at various scales, distinguishing between thick and thin vessels. The method integrates a dual-path network that employs convolution and sampling operations for advanced feature learning, leading to efficient end-to-end segmentation. The local vessel segmentation network features an encoder-decoder structure that retains spatial dimensions and employs dilated convolution to enhance the precision of thin vessel segmentation. A skip connection is added to further refine the segmentation of small vessels. The results on the DRIVE and CHASE_DB1 datasets show that this method outperforms existing techniques, achieving higher accuracy, sensitivity, and specificity. It successfully segments small, low-contrast vessels that are often overlooked while preserving the integrity and connectivity of the vascular structure.

The change in morphology, diameter, branching pattern and/or tortuosity of retinal blood vessels is an important indicator of various clinical disorders of the eye and the body. This paper reports an automated method for segmentation of blood vessels in retinal images by means of a unique combination of differential filtering and morphological processing. The centerlines are extracted by the application of first order derivative of Gaussian in four orientations and then the evaluation of derivative signs and average derivative values is made. The shape and orientation map of the blood vessel is obtained by applying a multidirectional morphological top-hat operator followed by bit plane slicing of a vessel enhanced grayscale image. The centerlines are combined with these maps to obtain the segmented vessel tree. The approach is tested on two publicly available databases and results show that the proposed algorithm can obtain robust and accurate vessel tracings with a performance comparable to other leading systems.

The low-contrast and narrow blood vessels in retinal images are difficult to be extracted but useful in revealing certain systemic disease. Motivated by the goals of improving detection of such vessels, we propose the radial projection method to locate the vessel centerlines. Then the supervised classification is used for extracting the major structures of vessels. The final segmentation is obtained by the union of the two types of vessels after removal schemes. Our approach is tested on the STARE database, the results demonstrate that our algorithm can yield better segmentation.

The problem of determining directions of blood vessels in the optic disk is considered. The proposed method for estimating the vessel directions is based on analyzing local minima of gray-scale profile of the eye-ground image. Results of tests on real images are presented.

Image analysis is a relevant tool to improve the healthcare services. Fundus retinal image analysis allows the early detection of ophthalmic diseases such as diabetes and glaucoma. Thus, growing interest is observed on the development of segmentation algorithms for blood vessels in retinal images. For this purpose, Kernel-based approaches with Gaussian matched filters have been successfully used. Nowadays, improved image sensors and computers deliver high resolution images, and different parameter values are required for the efficient operation of such filters. In this work, an optimization system using genetic algorithms is designed to calculate those values. To evaluate our methodology, a segmentation algorithm is proposed and the outcomes are evaluated on the HRF image database. Performance measures are obtained and compared to those obtained using state of the art methods. This analysis represents a first step in the detection and classification of normal and abnormal eye conditions.

In clinical medicine, the segmentation of blood vessels in retinal images is essential for subsequent analysis in clinical diagnosis. However, retinal images are often noisy and their vascular structure is relatively tiny, which poses significant challenges for vessel segmentation. To improve the performance of vessel segmentation, an improved model IterNet++ based on the architecture of IterNet is proposed. First, curvelet signal analysis is applied to enhance retinal images. Second, residual convolution (ResConv) blocks and guided filters are introduced to utilise the encoder features of previous iterations in the model to reduce overfitting. Third, offline hard‐sample mining is used to improve segmentation performance by utilising training samples with low segmentation accuracy as many possible on a few‐sample training set. In addition, a test‐time augmentation method is applied to testing samples in test dataset during inference. Extensive experiments show that this model achieves Dice scores of 0.8313, 0.8277, and 0.8372 on DRIVE, CHASE‐DB1, and STARE datasets, respectively, demonstrating the best performance compared with IterNet and other baseline models.