Printer Friendly

Segmentation algorithms for extraction of coronary vessels in angiographic images--a survey.

INTRODUCTION

Image segmentation is the process of partitioning the image into different areas based on the properties of the image. The segmentation process plays an important role medical image processing. Angiography is a well known technique used in medical image to find out the blocks in coronary artery vessels. There are various techniques available for segmentation. The algorithm which is suitable for one type of image may not be suitable for other kind of image. It is a critical task to find out, which algorithm works better angiography images. In this paper Section II lists the different segmentation techniques. Section III discusses the various algorithms which is used for blood vessel segmentation in angiography images. Section IV summarizes all the algorithms and its features and performance. Section V gives conclusion of the survey of different segmentation algorithms.

Image Segmentation Techniques:

Segmentation techniques are broadly classified into five categories based on the property and type of images dealt with. The categories are

1. Threshold based segmentation

2. Edge based segmentation

3. Region based segmentation

4. Clustering based segmentation

5. Texture based segmentation

A. Threshold based segmentation:

This technique is the simplest method for segmentation. It uses some threshold (pixel intensity) value to segment the images. The pixels whose intensity is above the threshold value are classified as one segment. The pixels whose intensity is below the threshold value are classified as another segment. The thresholding may be global thresholding, variable thresholding or multiple thresholding.

Global thresholding: Only one threshold value for the entire image.

Variable thresholding: The threshold value changes over the image.

Multiple thresholding: More than one threshold value is used.

B. Edge based segmentation:

Edge is nothing but local change in the image intensity. Usually it occurs at boundary between two regions. This basic criterion is used as the key and the segmentation is done by edge detection. There are several methods available for edge detection. Some of them are Canny Edge Detection, Roberts edge detection, Prewitt edge detection, Sobel Edge Detection, Kirsh edge detection, Marr-Hildreth edge detection, Robinson edge detection and LoG edge detection.

C. Region based segmentation:

Region based segmentation is based on some common property of the image. There are two types.

Region merging: Merging starts from a seed point. Include the neighboring pixels to the region until pixel with different property is encountered.

Region splitting: Divide the images into subparts until the region with uniform property is obtained.

D. Clustering techniques:

Clustering is a method of grouping the pixels based on the common feature of pixels. Various categories of clustering are

Partitioning-based clustering: These algorithms forms the clusters at once based on the property of the image.

Hierarchical clustering: these algorithms forms the clusters based on the previous clusters.

Density-Based Methods: In these methods clusters are formed based on the density of the object. If it exceeds some threshold then the pixels are grouped as cluster.

E. Texture based methods:

Texture is a kind of pattern which was derived from the similarity of the image information. The pattern may depend on the arrangement of pixel or intensity of pixel. It helps the user for the segmentation of images based on the texture property. The texture in image can be identified in any one of the two ways either the structured approach or statistical approach.

Review Of Literature:

This section discusses various segmentation algorithms under each category. It also briefs about the methodology and performance factors in each algorithms.

A. Threshold based models:

Marco Boegel et al. [5] proposed a method which is based on global thresholding and local adaptive segmentation techniques. It gives the accuracy of 98% in case of larger vessels and 89% in case of smaller vessels.

M. Ghalehnovi et al. [11] presents a fuzzy level set method for the segmentation of coronary vessels in 2D X-Ray angiography. Initial contour is formed by fuzzy clustering and thresholding. Segmentation is done by LMF(Local Morphology Fitting) level set method.

Yuxi Lian et al. [12] proposed a global and iterative local thresholding for the nidus and vessel segmentation in DSA images. The original image is divided into number of subimages. Using Otsu's method pixels in each subimage are classified into two classes based on threshold. Then using the variance of the subimages, the third class of pixels are formed, which is again classified using iterative process.

B. Region based methods:

Shan Wang et al. [8] proposed a two step method for segmentation of coronary angiograms. In the first step image enhancement is done by multi-scale filtering technique using Hessian matrix. Then the second stage involves the use of region growing algorithm for the extraction of vessels from the enhanced image. This method has the advantage of extracting small and distal vessels because of multi-scale filtering technique.

Daniel S.D. Lara1 et al [1] presents a method for segmentation of the coronary vessels in angiograms. This paper discusses combines approach of region growing and differential geometry for segmentation. This method gives overall accuracy of 90% in identifying the coronary vessels. The processing flow of this method is give in Fig. 1

Kaiqiong Sun et al. [10] proposed a method to segment the vessels using local region based active contour model with morphological fitting energy. This method is also suitable for the images with intensity inhomogeneity. Preprocessing and postprocessing are not required for this method. This method is suitable for different kind of vessels like single vessel, branch vessels and vessels with different dimensions.

Yin Wang et.al. [4] presents an segmentation method for 3D CT angiograms based on active-contour method which considers both global and local intensity information for energy calculation. Bayesian probabilistic framework is used for the implementation of this method. Segmentation results are computed for both synthetic and real images. Four metrics are used for comparison, True positive(TP), False Positive(FP), False Negative(FN) and Overlapping metric(OM). The method proposed here gives the better performance compared to the existing methods.

Maryam Taghizadeh Dehkordi et al. [2] proposed an active contour model for vessel segmentation in angiograms. Image enhancement is performed using the directional Hessian-based filter. The segmentation is done by local feature fitting model which is based on energy function. This algorithm is tested on synthetic and actual images. The performance of this model is estimated by true positive fraction(TPF) and false positive fraction(FPF).The algorithms performance is relatively high when compared to the other modes like local binary fitting(LBF), Local Image Fitting(LIF) and local morphology fitting(LMF).

C. Morphology based model:

Sohini Roychowdhury et.al. [3] presents a three stage segmentation method. This method is applied for blood vessel segmentation of retina images. The image is preprocessed and two binary images are extracted out of the input image. One binary image is obtained by high pass filtering. Another binary image is obtained by morphological reconstruction. The regions common to both the images are classified as blood vessels. The proposed algorithm achieves an accuracy of 95.2% in vessel segmentation.

Harsha P. Jawale et al. [6] presents a segmentation method using morphological operations. Initially the image is preprocessed with Top hat operator which is used to enhance the contrast of the image. The preprocessed image is segmented by morphological close operation. After segmentation the coronary vessels are extracted by thresholding technique. To evaluate the performance of the algorithm Jaccard index is used. Lowest Jaccard Index of 0.077 is obtained using this method.

Masoomeh Ashoorirad et al. [9] proposed a method which involves Preprocessing, Fuzzy logic and morphological operations. Four linear filters are used such as horizontal sobel filter, vertical sobel filter, high pass filter and low-pass filter. These filters are applied to the Input image and four binary images are generated. These images are given as input to Fuzzy Inference System. Based on the Fuzzy inference rules the pixels are classified as vascular or non-vascular. Then applying the morphological operators the vessels are extracted. The segmentation process is illustrated in Fig. 2

D. Clustering based models:

Meng Li, Huiguang He et al. [7] presents a method to detect the coronary vessels in CT angiograms. Otsu threshold and superpixel clustering method is used for image reconstruction. Then graphcut segmentation method is used to detect the boundaries of vessels. Top hat filtering method is used to extract the vessels from the image. The method described here works well if the images are having enough contrast in the region of interest i.e. vessels.

Mahmoud Ramze Rezaee et al [14] proposed Pyramidal Segmentation and Fuzzy clustering which gives the correlation coefficient of 0.94 which is better compared to 0.79 obtained by fuzzy c-means algorithm without pyramidal segmentation.

R. GeethaRamani et al. [13] proposed a method which combines image processing and data mining techniques. The preprocessing step involves image cropping, color transformation, color channel extraction, contrast enhancement, gabor filtering and halfwave rectification. A feature vector is obtained as output of preprocessing. Principle component analysis is performed on feature vector. Then K-means clustering is used to extract the blood vessels. The segmented image is post processed by morphological operations. This method is validated on DRIVE database and it gives an accuracy of 95%.

RESULTS AND DISCUSSION

Conclusion:

In this paper the segmentation methods based on different techniques are presented. The parameters for performance analysis true positive, false positive, false negative, overlapping metric, jaccard index are discussed. From the survey it is clear that the morphological based segmentation gives more accuracy in case of coronary vessels. Threshold based algorithms are not able to find out the tiny vessels. Region based methods uses hessian matrix and contour model which is complex in nature. Clustering based methods gives better results in case of retinal images.

REFERENCES

[1.] Daniel, S.D. Lara, Alexandre W.C. Faria, Arnaldo de A. Araujol, and D. Menotti, 2013. "A Novel Hybrid Method For The Segmentation Of The Coronary Artery Tree in 2D Angiograms", International Journal of Computer Science & Information Technology (IJCSIT) 5: 3.

[2.] Maryam Taghizadeh Dehkordi, Ali Mohamad Doost Hoseini, Saeed Sadri, Hamid Soltanianzadeh, 2014. "Local feature fitting active contour for segmenting vessels in angiograms", IET Computer Vision, 8(3): 161-170.

[3.] Sohini Roychowdhury, Dara D. Koozekanani, Keshab K. Parhi, 2015. "Blood Vessel Segmentation of Fundus Images by Major Vessel Extraction and Subimage Classification", IEEE Journal Of Biomedical And Health Informatics, 19: 3.

[4.] Yin Wang and Panos Liatsis, 2012. "Automatic Segmentation of coronary Arteries in CT Imaging in the Presence of Kissing Vessel Artifacts", IEEE Transactions On Information Technology In Biomedicine, 16: 4.

[5.] Marco Boegel, Philip Hoelter, Thomas Redel, Andreas Maier, Joachim Hornegger and Arnd Doerfler, 2015. "A Fully-Automatic Locally Adaptive Thresholding Algorithm for Blood Vessel Segmentation in 3D Digital Subtraction Angiography", IEEE Conference.

[6.] Harsha, P. Jawale, Prof. K.S. Bhagat, 2016. "Blood vessel segmentation in coronary angiogram image", IJEDR, 4: 3.

[7.] Meng Li, Huiguang He*, Jianhua Yi, Bin Lv, 2009. "Segmentation and tracking of coronary artery using graph-cut in CT angiographic", 2nd Inernaional Conference on Biomedical Engineering and Informatics, 2009. BMEI '09.

[8.] Shan Wang, Shenzhen, 2012. "A Segmentation method of Coronary Angiograms based on Multi-scale Filtering and Region-Growing", Inernaional Conference on Biomedical Engineering and Biotechnology (iCBEB).

[9.] Masoomeh Ashoorirad, Rasool Baghbani, 2009. "Blood Vessel Segmentation in Angiograms using Fuzzy Inference System and Mathematical Morphology", International Conference on Signal Processing Systems.

[10.] Kaiqiong Sun, Zhen Chen, and Shaofeng Jiang, 2012, "Local Morphology Fitting Active Contour for Automatic Vascular Segmentation", IEEE Transactions On Biomedical Engineering, 59: 2.

[11.] Ghalehnovi, M., E. Zahedi, E. Fatemizadeh, 2014. "Integration of Spatial Fuzzy Clustering with Level Set for Segmentation of 2-D Angiogram", IEEE Conference on Biomedical Engineering and Sciences, 8-10.

[12.] Yuxi Lian, Yuanyuan Wang, Jinhua Yu, Yi Guo, and Liang Chen, 2015. "Segmentation of Arteriovenous Malformations Nidus and Vessel in Digital Subtraction Angiography Images Based on an Iterative Thresholding Method", 8th International Conference on BioMedical Engineering and Informatics.

[13.] GeethaRamani, R., Lakshmi Balasubramanian, 2016. "Retinal blood vessel segmentation employing image processing and data mining techniques for computerized retinal image analysis", Biocybernetics and Biomedical Engineering 36, 102-118, ELSEVIER

[14.] Mahmoud Ramze Rezaee, Pieter M.J. van der Zwet, Boudewijn P.F. Lelieveldt, Rob J. van der Geest, and Johan H.C. Reiber, 2000. "A Multiresolution Image Segmentation Technique Based on Pyramidal Segmentation and Fuzzy Clustering", IEEE Transactions On Image Processing, 9: 7.

(1) Sambath. M and (2) Dr. D. John aravindhar

(1) Sambath.M, Research Scholar, Department of Computer Science & Engineering, Hindustan University, Chennai, Tamiinadu, INDIA.

(2) Dr. D.John aravindhar, Associate Professor, Department of Computer Science & Engineering, Hindustan University, Chennai, Tamiinadu, INDIA.

Received 12 February 2017; accepted 20 March 2017; published 25 March 2017

Address For Correspondence:

Sambath.M, Research Scholar, Department of Computer Science & Engineering, Hindustan University, Chennai, Tamilnadu, INDIA.

E-mail: msambath@hindustanuniv.ac.in

Caption: Fig. 1: Segmentation of Coronary artery

Caption: Fig. 2: Segmentation by Morphology
Table 1: Summary of different segmentation algorithms

Segmentation    Algorithm proposed by        Remarks
Method

Threshold       Marco Boegel et. al. [5]      89% accuracy is
based model                                  achieved.

                M. Ghalehnovi et al. [11]     Suitable for 2D X-ray
                                             angiography.

                Yuxi Lian et al. [12]         Suitable for DSA
                                             images.

Region          Shan Wang et al. [8]          Small and distal
based models                                 vessels also be
                                             extracted..

                Daniel S.D. Lara1 et.al[1]   90% accuracy is
                                             achieved.

                Kaiqiong Sun et al. [10]      Also works well for
                                             images with intensity
                                             inhomogenity.

                Yin Wang et.al. [4]           Suitable for 3D CT
                                             angiograms.

                Maryam Taghizadeh            works well for
                Dehkordi et al. [2]          images with intensity
                                             inhomogenity.

Morphology      Sohini Roychowdhury          Suitable for fundus
based models    et.al. [3]                    images.

                Harsha P. Jawale et a/. [6]   Lowest jaccard index of
                                             0.077 is achieved.

                Masoomeh Ashoorirad et al.   Minute vessels can be
                [9]                          extracted.

Clustering      Meng Li, Huiguang            Blood vessels are
based models    He et al. [7]                 tracked and changes are
                                             detected.

                Mahmoud Ramze Rezaee et      Better correlation
                al. [14]                      coefficient of 0.94

                R. GeethaRamani et al. [13]   Effective for retinal
                                             images
COPYRIGHT 2017 American-Eurasian Network for Scientific Information
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2017 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Aravindhar, D. John; Sambath. M.
Publication:Advances in Natural and Applied Sciences
Article Type:Report
Geographic Code:9INDI
Date:May 1, 2017
Words:2314
Previous Article:Hardware implementation of interleaved boost converter for wind-solar hybrid system.
Next Article:Maximum power point tracking based stand alone photovoltaic system using PSO algorithm.
Topics:

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |