Ultrasound image segmentation with multilevel threshold based on differential search algorithm

IET Image ProcessingVolume 13, Issue 6 p. 998-1005 Research ArticleFree Access Ultrasound image segmentation with multilevel threshold based on differential search algorithm Dangguo Shao, Dangguo Shao Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this authorChunrong Xu, Chunrong Xu Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this authorYan Xiang, Corresponding Author Yan Xiang 50691012@qq.com Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this authorPeng Gui, Peng Gui School of Computer Science and Technology, Wuhan University, Wuhan, People's Republic of ChinaSearch for more papers by this authorXiaofang Zhu, Xiaofang Zhu Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this authorChao Zhang, Chao Zhang Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this authorZhengtao Yu, Zhengtao Yu Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this author Dangguo Shao, Dangguo Shao Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this authorChunrong Xu, Chunrong Xu Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this authorYan Xiang, Corresponding Author Yan Xiang 50691012@qq.com Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this authorPeng Gui, Peng Gui School of Computer Science and Technology, Wuhan University, Wuhan, People's Republic of ChinaSearch for more papers by this authorXiaofang Zhu, Xiaofang Zhu Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this authorChao Zhang, Chao Zhang Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this authorZhengtao Yu, Zhengtao Yu Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunmingn, Yunnan, People's Republic of ChinaSearch for more papers by this author First published: 03 April 2019 https://doi.org/10.1049/iet-ipr.2018.6150Citations: 5AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract Ultrasound (US) image segmentation plays a very important role in diagnostic imaging. In order to extract special tissues from images, this study proposes a new method for segmentation of US images. The proposed method uses multilevel threshold segmentation which is based on Otsu and differential search algorithm. Testing in simulation US images shows that the proposed algorithm has a better result of segmentation than the three existing methods, including region growing, the active contour model and k-means technique. The proposed method gets the highest Fm values and the smallest area errors in experiments. Vivo US images are also tested by the proposed method and it achieves a good segmentation result. 1 Introduction Medical ultrasound (US) is one of medical imaging included US image, computed tomography (CT), magnetic resonance imaging and positron emission CT. The improvement of the US image processing is significant for the clinical diagnosis. In recent years, many researches have been made in noise reduction, texture analysis and classification [1-3]. Image segmentation is one of the important preprocessing for medical diagnosis and guidance. However, the low contrast of the images and fuzziness of the edge caused by speckle usually make the segmentation to be more difficult. There are many methods for image segmentation, such as threshold segmentation, region growing, active contour model, k-means and graphics cut. Due to its simple and effective property, the threshold method has become one of the most widely used methods [4-12]. Otsu proposed a threshold selection based on a maximum between-class variance [4]. Patra et al. [5] presented the energy curve analysis as an aid in threshold selection. Pun and Pun [6] proposed a threshold selection based on the entropy of histogram. For medical image segmentation, threshold segmentation is also an effective method. When the image has multiple targets or noise, one threshold segmentation may not be suitable. Region growing is widely used in common image or medical image segmentation [13]. It is necessary for traditional regional growing to select the seeds and parameters in the segmentation. Recently, researchers have taken the effort in the region growing improvement, such as automatic seed selection and the adaptive parameter [14, 15]. The active contour model is widely used in medical imaging segmentation [16]. Snake [17], GVF [18] and Chan and Vese [19] are often used in active contour model. Researchers also have applied the active contour model to the medical image include US image segmentation [20, 21]. Cluster segmentation is a very important and widely used segmentation technique in the field of medical image segmentation. K-means is one of the clustering methods [22, 23]. In recent years, particular attention has been paid to multi-threshold segmentation in medical image segmentation [7, 24-26]. Zhou et al. [24] present a novel hybrid bat algorithm for the multilevel thresholding in medical image segmentation. Zhang et al. [25] proposed an US image segmentation based on multi-scale fuzzy C-means method integrated with particle swarm optimisation (PSO). After the analysis of the experimental results, the threshold segmentation method proved to be a more suitable method for the noise and fuzzy edge features of US images. The Otsu method is an unsupervised method of automatic threshold selection for image segmentation. We used multi-threshold selection based on Otsu for US image segmentation and found that the results were appropriate. However, when the number of threshold increases, the exhaustive search strategy cannot obtain the thresholds quickly. Therefore, in recent years, researchers have tried to use search strategies, such as PSO [27], differential evaluation [28, 29], firefly algorithm [30], cuckoo search [31] and bat algorithm [32] to search for the optimal value of the objective function. In this paper, a new and efficient search algorithm, differential search (DS) algorithm, is taken into consideration [33]. DS is fast and accurate in searching for the best value and is very helpful for the segmentation of US images in this method. The proposed method often used for common image segmentation, but is rarely used for US images. The remaining part of this paper is organised as follows. Section 2 describes the Otsu threshold method. The segmentation algorithm and the criteria for quantifying performance are illustrated in Section 3. The material and the experiment platform, evaluation methods are demonstrated in Section 4, also, the quantitative quality metrics are applied and discussed in this section. The paper is concluded and potential future directions are presented in Section 5. 2 Otsu threshold Let L be intensity levels in image and these levels are in the range [4]. Then, the normalised histogram can be defined as follows: (1)where i represents a specific grey level and hi denotes the number of pixels for the corresponding grey level i. In the bi-level thresholding [11, 12], a threshold T = t divides the given image into two classes C1 and C2. The class C1 includes the grey levels in the range [0, t − 1], and class C2 contains the grey levels in the range [t, L − 1]. The probability distributions for the grey levels C1 and C2 are expressed as follows: (2)The mean levels m1 and m2 for C1 and C2 can be defined as follows: (3)The mean intensity of the entire image can be represented as follows: (4)The between-class variance function can be defined as follows: (5)Obtaining the optimal threshold is the one that maximises the objective function (6)The bi-level threshold segmentation can be extended to multilevel threshold segmentation. Let us assume that the image is divided into k classes (C1, C2, …, Ck) with k − 1 thresholds [t1, t2, …, tk−1]. Here, C1 includes the grey levels in the range [0, t1 − 1], while C2 includes the grey levels in the range [t1, t2−1], …, and Ck include the grey levels in the range [tk−1, L − 1], then it can be defined by (7)where Pk and mk are given by (8) (9)where is given by (4). Obtaining the thresholds , , …, by maximising the function of the variance between classes (10) 3 Segmentation algorithm 3.1 DS algorithm To search the best thresholds of the objective function more efficiently, we introduce the DS algorithm. DS algorithm simulates the Brownian-like random-walk movement [33]. The artificial-organisms (Xi, i = {1, 2, 3, …, N}) make up an artificial-organism (i.e. Superorganismg, g = 1, 2, 3, …, max generation) contain members as much as the size of the problem (i.e. Xij, j = {1, 2, 3, …, D}). Here, N is the number of elements in the superorganism and D indicates the dimension the respective problem. The mechanism of DS is as follows. First, initialise the artificial-organism (11)where upj and lowj represent the upper limit value and the lower limit value of the specific problem, respectively. In this paper, lowj and upj are 0 and 255, respectively. Then the artificial-organisms and the artificial-superorganism are defined as follows: (12) (13)Secondly, evaluate the fertility of each position. It needs an evaluation function to evaluate the fertility of the position. In this paper, the evaluation function is used (10). Thirdly, change the position of superorganism. A stopover site position is produced by using (14)The scale control, the size of the change occurred in the positions of the members of the artificial-organisms. The scale value is produced by using a gamma-random number generator controlled by a uniform-random number generator working in the range of [0,1] together. The scale can be defined as follows: (15)where R is a gamma-random number and mapi is a uniform-random number. The structure used for calculation of the scale value allows the respective artificial-superorganism to change direction radically in the habitat. The donor is randomly selected from the artificial-organisms Xi moving towards the targets. It is very important for a successful migration. If one element of the artificial-organisms goes beyond the limits of the habitat (i.e. search space), the element needs to be designated randomly to another position in the habitat. Fourthly, choose the most fertile stopover. After evaluating the stopover and comparing with the last position, the individual move to the stopover if the position of stopover is more fertile, or it remains in the source instead. The artificial organisms continue to migrate to the most fertile position until the migration stop. 3.2 Segmentation based on Otsu-DS In this paper, DS is used to search for thresholds that maximise the function of the variance between classes. The threshold is considered as individual of superorganism. Finding the best value of the between classes variance function can be seen as superorganism looking for the most fertile position. Then the processing of the segmentation method is as follows: Step 1: Preprocess image. The US image is filtered by bilateral filter. In order to achieve better segmentation results, we properly filter the image [1]. Bilateral filtering is a non-linear filter which can achieve the effect of edge protection and noise reduction. Step 2: Obtain the threshold by the proposed method (Otsu-DS). The threshold is the value which maximised the objective function (i.e. the fitness function) of the image to be divided, and is searched by the DS algorithm. Step 3: Segment the image by the multilevel threshold. In this paper, we use two thresholds to segment the US image. From the experimental tests, it is better to segment the US image into three classes with two thresholds. Also if there is one type of target in the US image, an optimal threshold is chosen from two thresholds. The criterion of selecting the better threshold is according to the intensity value of the target. If the target area has a large mean value, a larger threshold is selected; otherwise, a smaller threshold is selected. This threshold divides the image into two parts: target and background, as shown in Fig. 1. As the multiple targets in images, we use the thresholds to segment the image into multiple parts. Fig. 1Open in figure viewerPowerPoint Process of segmentation (a) Simulated US image, (b) Segment the image with two thresholds, (c) Dividing the image into binary image by one of the thresholds 3.3 Criteria for quantifying performance To assess the segmentation accuracy quantitatively, the Error and F-measure are utilised. Error is used to measure the misclassification rate, which is defined as follows [34]: (16)The F-measure is a criteria for quantifying the performance of segmentation results. It is defined as follows [35]: (17) (18) (19)where Rthreshold denotes the region segmented by the proposed method and Rideal denotes the ideal segmentation region. If the segmentation effect is more accurate, the Fm, precision and recall should be more larger. 4 Experiments and discussion 4.1 Material and platform The simulated US images are formed by convolving random scatterers with the point spread function (PSF) [36]. Simulated PSF is a separable 2D Gabor function consisting of the product of the 1D Gabor pulse and the 1D Gaussian radiation pattern. The impulse-excited pulse has a centre frequency of 7.5 and 6 MHz FWHM bandwidth. The −20 dB Gaussian apodisation (β = 0.1) is used. 100,000 random scatterers are distributed uniformly in this space with magnitude drawn from a uniform distribution. Sound speed is taken as 1540 m/s, and consequently wavelength λ = 0.0154 cm. The 2D Gaussian is sampled axially at system sampling frequency 40 MHz. A 60 dB additive white Gaussian noise is then added to the signal. The RF signal is converted to its complex analytic representation and demodulated into a base-band signal. We have verified our proposed algorithms using both simulated images and the vivo images. The vivo images obtained from Saset Healthcare iMago C21, a commercial Digital US scanner. MATLAB 2013b on a core (TM) i3-3120 personal computer, 2.50 GHz, 4 GB RAM is used as an experimental platform for all the experiments. 4.2 Results One cyst simulated US image and three cysts are formed by convolving random scatterers with the PSF, which are shown in Figs. 2a and 3a, respectively. Figs. 2 and 3 display the simulated US image and the result of the experiment. In order to verify the performance, the proposed method is compared with the region growing, active contour and k-means. In region growing, the four-neighborhood domain and the seed in region of interest (ROI) are selected for segmenting. The similarity distance of pixel intensity value is less than 0.05. In the active contour, the initial contour around the ROI is given and the iteration number is set to 500 for detecting the edge. In k-means method, the image is also grouped into three clusters and choose one of the clusters to segment the image as well. To verify the image segmentation capability of the proposed method, we also segmented simulated the image with light and dark regions in Fig. 4. The light region and dark region denote the two different tissues. Fig. 2Open in figure viewerPowerPoint Simulated US image of one cyst and the result of experiment (a) Simulated US image, (b) Ideal region. Segmented simulated US image using, (c) Region growing, (d) Active contour, (e) K-means, (f) Proposed method Fig. 3Open in figure viewerPowerPoint Simulated US image of three cysts and the result of experiment (a) Simulated US image, (b) Ideal region. Segmented simulated US image using, (c) Region growing, (d) Active contour, (e) K-means, (f) Proposed method Fig. 4Open in figure viewerPowerPoint Simulated image with light and dark regions and the result of experiment (a) Original image, (b) Ideal region, (c) Segmented by region growing, (d) Segmented by active contour, (e) Segmented by k-means, (f) Segmented by proposed method Table 1 shows the thresholds, which are searched by DS. We compare them with thresholds obtained by the exhaustive search. The exhaustive search gives all the threshold combinations suitable for the image, such as (1, 2), (1, 3), …,(244.255) and obtains the functional fitness of each combination. Select the threshold combination which obtained the best fitness. Exhaustive search can get thresholds accurately but not quickly. The quantifying performance of the methods are displayed in Tables 2–4. From the results, we found that the proposed method can search suitable thresholds to segment the image and extract the regions. The extract region are similar to the ideal region. Especially, we can see that the proposed method can extract the light and dark regions from the whole image more integrally. Quantitatively, the search results of the DS algorithm are as accurate as the exhaustive search method from Table 1. From Tables 2–4, the proposed method gets the highest Fm values and the smallest area errors in each experiment. It also takes less time than the region growing and the active contour model. In each experiment, the processing time of region growth and active contour is two to three times longer than that of the proposed method. Table 1. Comparison of fitness obtain by two different method Image Number of threshold Exhaustive search DS Threshold Fitness Threshold Fitness one cyst 2 (136,160) 367.1260 (136,160) 367.1260 three cysts 2 (139,161) 388.7295 (139,161) 388.7295 light and dark 2 (132,161) 165.9576 (132,161) 165.9576 Table 2. Comparison of segmentation results with one cyst Methods Precision Recall Fm Error Time, s region growing 0.9993 0.9456 0.9717 0.0162 5.148807 active contour 0.9965 0.9709 0.9835 0.0093 37.155285 k-means 0.9950 0.9713 0.9830 0.0096 1.344271 proposed method 0.9932 0.9784 0.9857 0.0080 2.749473 Table 3. Comparison of segmentation results with three cysts Methods Precision Recall Fm Error Time, s region growing 0.9865 0.9358 0.9605 0.0277 6.517449 active contour 0.9873 0.9517 0.9692 0.0214 18.219206 k-means 0.9590 0.9347 0.9467 0.0368 0.791435 proposed method 0.9743 0.9663 0.9703 0.0204 2.751216 Table 4. Comparison of segmentation results with light and dark regions Methods Precision Recall Fm Error Time, s region growing 0.6781 0.9844 0.8030 0.0629 13.429084 active contour 0.4987 0.1695 0.2530 0.5602 69.006458 k-means 0.9535 0.7964 0.8679 0.0558 1.976883 proposed method 0.9581 0.9154 0.9363 0.0252 3.648068 To test the capability of the proposed method in vivo US image segmentation, three vivo US images, such as image 1, image 2 and image 3 are segmented by the proposed method. In the experiments, we segmented image for three classes with two thresholds. The first threshold is selected to extract the first class. Figs. 5-7 display the segmentation result of the vivo images. Fig. 5Open in figure viewerPowerPoint Segmentation result of the vivo image 1 (a) Original image 1, (b) Segmented by the region growing, (c) Segmented by the active contour, (d) Segmented by the k-means, (e) Segmented by the proposed method Fig. 6Open in figure viewerPowerPoint Segmentation result of the vivo image 2 (a) The original image 2, (b) Segmented by the region growing, (c) Segmented by the active contour, (d) Segmented by the k-means, (e) Segmented by the proposed method Fig. 7Open in figure viewerPowerPoint Segmentation result of the vivo image 3 (a) The original image 3, (b) Segmented by the region growing, (c) Segmented by the active contour, (d) Segmented by the k-means, (e) Segmented by the proposed method 4.3 Discussion of results Otsu multi-threshold selection based on histogram statistics gets the thresholds suitably. The Otsu method is an unsupervised method which can select threshold adaptively for image segmentation. DS algorithm allowed exploring any dimension of the problem. Combining Otsu with DS algorithm can obtain appropriate thresholds quickly for fast segmentation. Inanition, the obtaining thresholds and the segmentation result are stable. From the result of simulated US image segmentation and quantitative analysis, the proposed method gets the highest Fm values and the smallest area errors in each experiment. It can achieve the better segmentation results than existing methods (region growing, active contour and k-means). Region growing segment the images by grouping pixels with similar attributes. Although it can segment and extract region, the noise and multiple targets may lead to the result over segmentation or incomplete segmentation. Additionally, it is difficult for region growing to select the seed in the image with multiple targets. The active contour model detects the edge based on techniques of Chan–Vese model. It needs to give the initial contour when extracting the edge and the initial contour setting may also affect the segmentation result. The active contour cannot segment the image with light and dark regions well and it will get a large error. In addition, the active contour model detects the main outline of the image, but may not be accurate in detecting edge details. The proposed method selects adaptively threshold and segment efficiently image. Significantly, it is more easier and efficient to segment the multiple targets than region growing and active contour. K-means method is sensitive to the selection of the initial centre, and it is easy to make the cluster centre fall into a local optimum and affect the correctness of clustering. Obviously, the proposed algorithm has an advantage in terms of segmenting the image and extract the special tissues while using less processing time. 5 Conclusions US image segmentation is significant in clinical diagnosis. To extract special tissues, this paper proposed a multilevel threshold based on DS algorithm. The experiments and quantitative data show the proposed method can get better results in extracting special tissue efficiently. In that sense, our algorithm has an advantage over conventional region growing, the active contour model and k-means. Future work involving more US images will be tested to fine tune algorithms and optimise thresholds to expand the application of picture processing. 6 Acknowledgments This work was supported by the China Postdoctoral Science Foundation (2016M592894XB), the National Natural Science Foundation of China (grant nos. 61741112 and 81560296) and the Natural Science Foundation of Yunnan Province (grant nos. 2017FB097 and 2017FB098). The authors thank Na Zheng and Shuo Wang (Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, Yunnan, China) for many valuable suggestions during the algorithm development and testing in simulated ultrasound image and in vivo data. They appreciate the valuable comments from other members of their department. 7 References 1Shao D., Liu P., and Liu Dong C.: ‘Characteristic matching-based adaptive fast bilateral filter for ultrasound speckle reduction’, Pattern Recognit. Lett., 2013, 34, (5), pp. 463– 469 2Lee M.S., Yen C.L., and Ueng S.K.: ‘Speckle reduction with edges preservation for ultrasound images: using function spaces approach’, IET Image Process., 2012, 6, (7), pp. 813– 821 3Raghesh Krishnan K., and Radhakrishnan S.: ‘Focal and diffused liver disease classification from ultrasound images based on isocontour segmentation’, IET Image Process., 2015, 9, (4), pp. 261– 270 4Ostu N.: ‘A threshold selection method from gray level histogram’, IEEE Trans. Syst. Man Cybern., 1979, 9, (1), pp. 62– 66 5Patra S., Gautam R., and Singla A.: ‘A novel context sensitive multilevel thresholding for image segmentation’, Appl. Soft Comput., 2014, 23, (23), pp. 122– 127 6Pun T., and Pun T.: ‘A new method for gray-level picture thresholding using the entropy of the histogram’, Signal Process., 1985, 2, (3), pp. 223– 237 7Li Y., Jiao L., and Shang R. et al: ‘Dynamic-context cooperative quantum-behaved particle swarm optimization based on multilevel thresholding applied to medical image segmentation’, Inf. Sci., 2015, 294, pp. 408– 422 8Jiang Y., Tsai P., and Hao Z. et al: ‘Automatic multilevel thresholding for image segmentation using stratified sampling and tabu search’, Soft Comput., 2015, 19, (9), pp. 2605– 2617 9Rajinikanth V., Raja N.S.M., and Satapathy S.C.: ‘Robust color image multi-thresholding using between-class variance and cuckoo search algorithm’, Adv. Intell. Syst. Comput., 2016, 433, pp. 379– 386 10Horng M.-H.: ‘Multilevel thresholding selection based on the artificial bee colony algorithm for image segmentation’, Expert Syst. Appl., 2011, 38, (11), pp. 114– 134 11Sri Madhava Raja N., Rajinikanth V., and Latha K.: ‘Otsu based optimal multilevel image thresholding using firefly algorithm’, Model. Simul. Eng., 2014, 2014, (2), pp. 1– 17 12Ghamisi P., Couceiro M.S., and Martins F.M.L. et al: ‘Multilevel image segmentation based on fractional-order darwinian particle swarm optimization’, IEEE Trans. Geosci. Remote, 2014, 52, (5), pp. 2382– 2394 13Adams R., and Bischof L.: ‘Seeded region growing’, IEEE Trans. Pattern Anal., 2002, 16, (6), pp. 641– 647 14Gill H.K., Girdhar A., and Kaur J.: ‘ Automatic region growing segmentation for medical ultrasound images’. Int. Conf. on Medical Imaging, Greater Noida, India, 2015, pp. 454– 457 15Wang W., Li J., and Jiang Y. et al: ‘ An automatic energy-based region growing method for ultrasound image segmentation’. IEEE Int. Conf. on Image Processing, Quebec City, Canada, 2015, pp. 1553– 1557 16Zhou S., Wang J., and Zhang S. et al: ‘Active contour model based on local and global intensity information for medical image segmentation’, Neurocomputing, 2016, 186, (c), pp. 107– 118 17Kass M., Witkin A., and Terzopoulos D.: ‘Snakes: active contour models’, Int. J. Comput. Vision, 1988, 1, (4), pp. 321– 331 18Xu C., and Prince J.L.: ‘Snakes, shapes, and gradient vector flow’, IEEE Trans. Image Process., 1998, 7, (3), pp. 359– 369 19Chan T.F., and Vese L.A.: ‘Active contours without edges’, IEEE Press, 2001, 10, (2), pp. 266– 277 20Yu H., Pattichis M.S., and Goens M.B.: ‘ Robust segmentation of freehand ultrasound image slices using gradient vector flow fast geometric active contours’. IEEE Southwest Symp. on Image Analysis and Interpretation, Denver, CO, USA, March 2006, pp. 115– 119 21Liu B., Cheng H.D., and Huang J. et al: ‘Probability density difference-based active contour for ultrasound image segmentation’, Pattern Recogn., 2010, 43, (6), pp. 2028– 2042 22Saranya P.K., Samundeeswari E.S., and Manavalan R.: ‘Breast ultrasound (BUS) image segmentation using regularized rough K-means (RRKM) clustering: a novel approach’, J. Med. Imag. Health In., 2017, 7, (1), pp. 94– 100 23Nor A., Samy A.S., and Umi K.: ‘Adaptive fuzzy moving k-means clustering algorithm for image segmentation’, IEEE Trans. Consum Electron., 2009, 55, (4), pp. 2145– 2153 24Zhou Y., Li L., and Ma M.: ‘A novel hybrid Bat algorithm for the multilevel thresholding medical image segmentation’, J. Med. Imag. Health In., 2015, 5, (8), pp. 1742– 1746 25Zhang Q., Huang C., and Li C. et al: ‘ Ultrasound image segmentation based on multi-scale fuzzy c-means and particle swarm optimization’. IET Int. Conf. on Information Science and Control Engineering, Shenzhen, China, 2014, vol. 1, p. 37 26Li A., Li Y., and Wang T. et al: ‘ Medical image segmentation based on maximum entropy multi-threshold segmentation optimized by improved cuckoo search algorithm’. Int. Congress on Image and Signal Processing, Shenyang, China, 2016, pp. 470– 475 27Ghamisi P., Couceiro M.S., and Benediktsson J.A. et al: ‘An efficient method for segmentation of images based on fractional calculus and natural selection’, Expert Syst. Appl., 2012, 39, (16), pp. 12407– 12417 28Storn R., and Price K.: ‘Differential evolution – a simple and efficient heuristic for global optimization over continuous spaces’, J. Global Optim., 1997, 11, (4), pp. 341– 359 29Soham S., and Swagatam D.: ‘Multilevel image thresholding based on 2D histogram and maximum Tsallis entropy – a differential evolution approach’, IEEE Trans. Image Process., 2013, 22, (12), pp. 4788– 4797 30Zhou C., Tian L., and Zhao H. et al: ‘ A method of two-dimensional Otsu image threshold segmentation based on improved firefly algorithm’. IEEE Int. Conf. on Cyber Technology in Automation, Shenyang, China, 2015, pp. 1420– 1424 31Ashish K., Vineet K., and Girish K.: ‘Cuckoo search algorithm and wind driven optimization based study of satellite image segmentation for multilevel thresholding using Kapur's entropy’, Expert Syst. Appl., 2014, 41, (7), pp. 3538– 3560 32Rajinikanth V., Aashiha J.P., and Atchaya A.: ‘Gray-level histogram based multilevel threshold selection with Bat algorithm’, Int. J. Comput. Appl., 2014, 93, (93), pp. 1– 8 33Civicioglu P.: ‘Transforming geocentric Cartesian coordinates to geodetic coordinates by using differential search algorithm’, Comput. Geosci., 2012, 46, (3), pp. 229– 247 34Xu Y.: ‘ A modified spatial fuzzy clustering method based on texture analysis for ultrasound image segmentation’. IEEE Int. Symp. on Industrial Electronics, Seoul, South Korea, 2009, pp. 746– 751 35Wang Z.: ‘A new approach for segmentation and quantification of cells or nanoparticles’, IEEE Trans. Ind. Inf., 2016, 12, (3), pp. 962– 971 36Liu P., and Liu D.C.: ‘Filter-based compounded delay estimation with application to strain imaging’, IEEE Trans. Ultrason. Ferroelectr., 2011, 58, (10), pp. 2078– 2095 Citing Literature Volume13, Issue6May 2019Pages 998-1005 FiguresReferencesRelatedInformation