Retinal blood vessel segmentation using pixel-based feature vector

https://doi.org/10.1016/j.bspc.2021.103053Get rights and content

Highlights

  • We have presented a pixel-based method for retinal vessel segmentation in this paper.

  • We used the classical edge detection methods effectively.

  • The proposed method is independent of the dataset.

  • Experimental results are satisfactory and robust.

Abstract

A lot of important disease information can be accessed by performing retinal blood vessel analysis on fundus images. Diabetic retinopathy is one of the diseases understood by retinal blood vessel analysis. If this disease is detected at an early stage, vision loss can be prevented. In this paper, a method that performs retinal blood vessel analysis with classical methods is proposed. In this proposed system, pixel-based feature extraction is performed. Five different feature groups are used for feature extraction. These feature groups are edge detection, morphological, statistical, gradient, and Hessian matrix. An 18-D feature vector is created for each pixel. This feature vector is given to the artificial neural network for training. Using test images, the system is tested on two publicly available datasets. Sensitivity, Specificity, and Accuracy performance measures were used as success measures. The similarity index between the segmented image and the ground truth is measure using Dice and Jaccard. The accuracy of the system was measured as 96.18% for DRIVE and 94.56% for STARE, respectively. Experimental results show that the proposed algorithm achieves satisfactory results. This method can be used as an automated retinal blood vessel segmenting system.

Introduction

Computer-aided systems (CAD) are widely used in the medical field. CADs are preferred because they give robust, accurate, and fast results. Also, CADs alone does the work of many experts. Thus, it minimizes errors caused by human negligence. One of the CADs is fundus image analysis. Using fundus images can be detected fields such as optic disc [1], [2], [3], [4], optic cup [5], [6], [7], [8], fovea [9], [10], [11], [12], macula [13], [14], exudate [3], [4], [15], [16], [17] and blood vessels [18], [19], [20], [21], [22].

Automated segmentation of retinal blood vessels is an important step for computer-aided fundus image analysis. Many diseases such as age-related macular degeneration [23], [24], glaucoma [25], [26], diabetes [27], [28], [29], [30], [31] and hypertension [32], [33], [34] can be detected using the retinal blood vessel in the fundus image. With the increasing number of these diseases, the variety of retinal vascular segmentation methods has also increased.

In literature are generally presented various computer-aided methods for diseases by using features such as the branching angle, width of the retinal vessel, and the calibrated curvature of the retinal vessel. Beaudelaire et al. [35] proposed a vessels segmentation method based on the classical edge detection filters and artificial neural networks. In their method, they used edge detection filters to extract the feature vector. The extracted features are given as an introduction to an artificial neural network to understand whether each pixel belongs to blood vessels. Orujov et al. [36] enhanced an image processing algorithm for the detection of blood vessels in retinal fundus images. In this algorithm, Mamdani (Type-2) fuzzy rules are used. Jebaseeli et al. [37] proposed the system for retinal blood vessel segmentation. The proposed system uses Tandem Pulse Coupled Neural Network (TPCNN) model for automatic feature vector creation and then Deep Learning-Based Support Vector Machine (DLBSVM) for classification and extraction. The TPCNN model operates on inter and intra channel linking of the input neurons. Wang et al. [38] proposed a new cascade classification framework for retinal vessel segmentation. Vessel segmentation consists of Mahalanobis distance classifiers to form an extremely nonlinear decision Mahalanobis distance is computationally efficient. Here, network size is not defined in advance but is determined by the complexity of the training data. Ramos et al. [39] reduce the noise of the green channel of the original image using a Low-Pass Radius Filter. Then, a 30-element Gabor filter and a Gaussian fractional derivative are used to notably enhance the vascular segmentation. Finally, a threshold method and morphological process are applied to isolate the vascular structure. The method proposed by Zhu et al. [40] is based on Extreme Learning Machine (ELM) for segmentation of retinal vessels. In the training step, a feature vector is created for each pixel of the training retinal image. This feature vector consists of morphological progress, Hessian matrix, local intensity, and divergence of vector fields. To obtain a classifier is used the feature vector and manual labels. Rezaee et al. [41] suggested an efficient method based on Fuzzy Entropy. Firstly, the blurring noises are eliminated with a Wiener filter provided. Secondly, the adaptive filter is used and blood vessels are extracted at a basic level. Rodrigues et al. [42] proposed a method using wavelets, Hessian, and morphology. In this method, both optical disc detection and retinal vascular network structure were determined. Researchers worked on the tubular characteristic of the retinal blood vessel to determine veins and arteries. Cuevas et al. [43] implemented a method called Lateral Inhibition (LI) on the retinal images. LI is a technique inspired by nature. The author's goal is to improve the contrast between the retinal vessels and the background of the image. Also, it is decided whether an image pixel belongs to the blood vessel by finding a suitable threshold value. For the threshold value, the Differential Evolution (DE) algorithm based on cross-entropy is used. Khan et al. [44] improved an effective method for automatically extracting blood vessels from color retinal images. This method relies on an edge operator. Edge operator is called the name normalized Gaussian Derivative Kernel in the second order. Principal component analysis-based image preprocessing is applied to the images before this operator is applied to the images. Sigursson et al. [45] presented a method based on feature extraction for the detection of blood vessels in the retinal images. This presented method focused on two vessel features. The first feature is extracted using the local minimum detection method. The second feature is extracted using the edge detection method. Nayebifar et al. [46] proposed a novel method that relies on particle filters. The purpose of the authors is to determine and locally track the vessel paths in the retinal image. The method uses a probability density function (PDF). Firstly, optic disc localization is determined. Then a recursive tracking process begins using an appropriate set of starting points. Chaudhurı et al. [47] presented an MF filter based on optical and spatial properties of vessels to be recognized. MF was built on assumption. In this assumption, the gray-level profile of the cross-section of the vessel is thought to be the shape of Gaussian. The authors created a set of 12 templates that are used to match the vessels along with the different directions. However, MF may vary in many ways rely on the parameters. Al-Rawi et al. [48] optimized the parameters of MF using a genetic algorithm. Cinsdikici et al. [49] optimized the parameters of MF using an ant colony algorithm. Sreejini et al. [50] optimized the parameters of MF using a particle swarm optimization algorithm.

One of the main problems of blood vessel segmentation in retinal fundus images is poor image quality. Thus, in literature generally applied enhancement methods before retinal vessel segmentation were performed. In this paper, the contrast limited adaptive histogram equalization (CLAHE) method is used in the preprocessing stage. The CLAHE method is one of the most popular preprocessing methods used on retinal fundus images [51], [52], [53], [54], [55]. A median filter was applied to images after the CLAHE method. The Median filter is used to reduce other noise on the preprocessed image. After the preprocessing stage, the feature vector generation stage begins. The feature vector is created by subtracting 18-D features from each pixel of the preprocessed image. The data set is trained using these features. Artificial Neural Network (ANN) is used in the training stage. After the training stage, the test process starts. Thus, the pixel-based retinal vessel segmentation process is realized. In the proposed method, an accuracy rate of 0.9618 was obtained for the DRIVE data set, and an accuracy rate of 0.9456 was obtained for the STARE data set.

This paper focuses on using classic features effectively. The critical contributions are as follows:

• Instead of all traditional edge detection methods, this paper presents these methods combined as a single feature.

• The statistical properties are presented in a different technique to increase the robustness of the vessel segmentation.

• The proposed method was built independently from the database.

The organization of the rest of this paper is as follows: Material and details of the method are presented in Section 2. Experimental results and discussion of the proposed method are given in Section 3. Finally, the conclusion is given in Section 4.

Section snippets

Materials and methods

The proposed method presents a vessel segmentation method by extracting pixel-based features on retinal images. In the proposed method, there is a preprocessing stage first. A pixel-based feature vector has been extracted on the preprocessed images. The created feature vector has been given as an input to the ANN. Testing is performed after the ANN is trained. The testing process has been carried out on public datasets. The created feature vector for each pixel is given in Table 1. The block

Experimental results and discussion

An 18-D feature vector is extracted from each pixel for retinal blood vessel segmentation. The feature vector is extracted from 5 different feature groups. Each group has effective knowledge of the characteristic feature of the retinal blood vessel pixel. The edge detection methods positively affected the success of the experimental results. These edge features were preferred because they produced successful results in other studies. [36], [58]. Statistical features were the feature group with

Conclusion

In this paper, a method for segmenting retinal blood vessels is proposed. The proposed method is based on pixel-based feature extraction. Five different feature groups are used for feature extraction. These feature groups are edge detection methods, morphological methods, statistical methods, gradient-based methods, and hessian-based methods. Edge detection algorithms are presented as a single feature, not separately. Windows are used while extracting pixel-based statistical features. Features

CRediT authorship contribution statement

Buket Toptaş: Software, Methodology. Davut Hanbay: Software, Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgement

This work was supported by the Inonu university scientific research and coordination unit [FDK-2020-2109]

References (76)

  • A. Fathi et al.

    Automatic wavelet-based retinal blood vessels segmentation and vessel diameter estimation

    Biomed. Signal Process. Control.

    (2013)
  • Y. Wang et al.

    Retinal vessel segmentation using multiwavelet kernels and multiscale hierarchical decomposition

    Pattern Recognit.

    (2013)
  • K. Narasimhan et al.

    Hypertensive retinopathy diagnosis from fundus images by estimation of AVR

    Procedia Eng.

    (2012)
  • B. Saha Tchinda et al.

    Retinal blood vessels segmentation using classical edge detection filters and the neural network

    Informatics Med. Unlocked.

    (2021)
  • T.J. Jebaseeli et al.

    Segmentation of retinal blood vessels from ophthalmologic Diabetic Retinopathy images

    Comput. Electr. Eng.

    (2019)
  • X. Wang et al.

    Blood vessel segmentation from fundus image by a cascade classification framework

    Pattern Recognit.

    (2019)
  • H. Aguirre-Ramos et al.

    Blood vessel segmentation in retinal fundus images using Gabor filters, fractional derivatives, and Expectation Maximization

    Appl. Math. Comput.

    (2018)
  • C. Zhu et al.

    Retinal vessel segmentation in colour fundus images using Extreme Learning Machine

    Comput. Med. Imaging Graph.

    (2017)
  • K. Rezaee et al.

    Optimized clinical segmentation of retinal blood vessels by using combination of adaptive filtering, fuzzy entropy and skeletonization

    Appl. Soft Comput. J.

    (2017)
  • L.C. Rodrigues et al.

    Segmentation of optic disc and blood vessels in retinal images using wavelets, mathematical morphology and Hessian-based multi-scale filtering

    Biomed. Signal Process. Control.

    (2017)
  • B. Nayebifar et al.

    A novel method for retinal vessel tracking using particle filters

    Comput. Biol. Med.

    (2013)
  • M. Al-Rawi et al.

    Genetic algorithm matched filter optimization for automated detection of blood vessels from digital retinal images

    Comput. Methods Programs Biomed.

    (2007)
  • M.G. Cinsdikici et al.

    Detection of blood vessels in ophthalmoscope images using MF/ant (matched filter/ant colony) algorithm

    Comput. Methods Programs Biomed.

    (2009)
  • K.S. Sreejini et al.

    Improved multiscale matched filter for retina vessel segmentation using PSO algorithm

    Egypt. Informatics J.

    (2015)
  • J. Dash et al.

    A thresholding based technique to extract retinal blood vessels from fundus images

    Futur. Comput. Informatics J.

    (2017)
  • M.M. Fraz et al.

    An approach to localize the retinal blood vessels using bit planes and centerline detection

    Comput. Methods Programs Biomed.

    (2012)
  • G. Azzopardi et al.

    Trainable COSFIRE filters for vessel delineation with application to retinal images

    Med. Image Anal.

    (2015)
  • Ç. Sazak et al.

    The multiscale bowler-hat transform for blood vessel enhancement in retinal images

    Pattern Recognit.

    (2019)
  • M.A.U. Khan et al.

    Thin Vessel Detection and Thick Vessel Edge Enhancement to Boost Performance of Retinal Vessel Extraction Methods

    Procedia Comput. Sci.

    (2019)
  • A.K. Shukla et al.

    A fractional filter based efficient algorithm for retinal blood vessel segmentation

    Biomed. Signal Process. Control.

    (2020)
  • X. You et al.

    Segmentation of retinal blood vessels using the radial projection and semi-supervised approach

    Pattern Recognit.

    (2011)
  • S. Aslani et al.

    A new supervised retinal vessel segmentation method based on robust hybrid features

    Biomed. Signal Process. Control.

    (2016)
  • A.H. Al-Sharfaa et al.

    Localization of Optic Disk and Exudates Detection in Retinal Fundus Images

    J. Phys. Conf. Ser.

    (2021)
  • S.R. Rupanagudi, V.G. Bhat, B.K. Revana, J.G. Chandramouli, S. Devegowda, G. Darshan, K. Kumaran, N. Malepati, L....
  • S. Liu et al.

    Joint optic disc and cup segmentation using semi-supervised conditional GANs

    Comput. Biol. Med.

    (2019)
  • H.N. Veena et al.

    A novel optic disc and optic cup segmentation technique to diagnose glaucoma using deep learning convolutional neural network over retinal fundus images, J. King Saud Univ. - Comput

    Inf. Sci.

    (2021)
  • N.M. Tan, D.W.K. Wong, J. Liu, W.J. Ng, Z. Zhang, J.H. Lim, Z. Tan, Y. Tang, H. Li, S. Lu, T.Y. Wong, Automatic...
  • C. Huang et al.

    A new deep learning approach for the retinal hard exudates detection based on superpixel multi-feature extraction and patch-based CNN

    Neurocomputing.

    (2021)
  • Cited by (27)

    • Brain MRI high resolution image creation and segmentation with the new GAN method

      2023, Biomedical Signal Processing and Control
      Citation Excerpt :

      Classical image segmentation approaches, Thresholding [1], Clustering methods [2], Histogram-based methods [3], Edge detection [4], Region-growing methods [5], Partial differential equation-based methods [6], Graph partitioning methods [7], Watershed transformation [8] can be listed. Classical segmentation techniques are widely used [9–11]. When innovative segmentation techniques are examined, it is seen that convolutional neural networks [12–16] and generative adversarial networks [17–19] are used.

    • CSGNet: Cascade semantic guided net for retinal vessel segmentation

      2022, Biomedical Signal Processing and Control
      Citation Excerpt :

      These methods could be roughly divided into traditional and deep learning-based methods. Traditional methods are often based on hand-crafted features [8–13], such as color, shape, etc. For instance, Saleh et al. [10] utilized the color characteristic of retinal vessels to segment vessels via a combination of histogram equalization and threshold selection.

    • A multi-branch hierarchical attention network for medical target segmentation

      2022, Biomedical Signal Processing and Control
      Citation Excerpt :

      These semantic regions were further recognized to segment the breast tumors through a BoVWs feature based classifier. In addition, dictionary learning [18], wavelet transform [19], feature vector [20], and random forest [21] were also investigated for medical image segmentation. Generally, the handcrafted feature based methods are easy to be implemented and show promising processing efficiencies.

    View all citing articles on Scopus
    View full text