Assessment of despeckle filtering algorithms for segmentation of breast tumours from ultrasound images

Assessment of despeckle filtering algorithms for segmentation of breast tumours from ultrasound images

BBE 310 1–17 biocybernetics and biomedical engineering xxx (2018) xxx–xxx Available online at www.sciencedirect.com ScienceDirect journal homepage: ...

NAN Sizes 0 Downloads 0 Views

BBE 310 1–17 biocybernetics and biomedical engineering xxx (2018) xxx–xxx

Available online at www.sciencedirect.com

ScienceDirect journal homepage: www.elsevier.com/locate/bbe 1 2 3

Original Research Article

Assessment of despeckle filtering algorithms for segmentation of breast tumours from ultrasound images

4 5 6

7 8 9

Q1

Kriti a, Jitendra Virmani b,*, Ravinder Agarwal a a b

Thapar Institute of Engineering and Technology, Patiala, Punjab 147004, India CSIR-Central Scientific Instruments Organisation (CSIR-CSIO), Chandigarh 160030, India

article info

abstract

Article history:

In the present work, the breast ultrasound images are pre-processed with various despeckle

Received 2 February 2018

filtering algorithms to analyze the effect of despeckling on segmentation of benign and

Accepted 4 October 2018

malignant breast tumours from ultrasound images. The despeckle filtering algorithms are

Available online xxx

broadly classified into eight categories namely local statistics based filters, fuzzy filters, Fourier

Keywords:

and hybrid filters. Total 100 breast ultrasound images (40 benign and 60 malignant) are

Breast ultrasound

processed using 42 despeckle filtering algorithms. A despeckling filter is considered to be

filters, multiscale filters, non-linear iterative filters, total variation filters, non-local mean filters

Speckle noise

appropriate if it preserves edges and features/structures of the image. Edge preservation

Despeckling filters

capability of a despeckling filter is measured by beta metric (b) and feature/structure preserva-

Image quality metrics

tion capability is quantified using image quality index (IQI). It is observed that out of 42 filters, six

Edge preservation

filters namely Lee Sigma, FI, FB, HFB, BayesShrink and DPAD yield more clinically acceptable

Segmentation

images in terms of edge and feature/structure preservation. The qualitative assessment of these images has been done on the basis of grades provided by the participating experienced radiologist. The pre-processed images are then fed to a segmentation module for segmenting the benign or malignant tumours from ultrasound images. The performance assessment of segmentation algorithm has been done quantitatively using the Jaccard index. The results of both quantitative and qualitative assessment by the radiologist indicate that the DPAD despeckle filtering algorithm yields more clinically acceptable images and results in better segmentation of benign and malignant tumours from breast ultrasound images. © 2018 Nalecz Institute of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences. Published by Elsevier B.V. All rights reserved.

12 10 11 13 14 15

1.

16 17

In the field of medical diagnosis ultrasound is widely used because of its non-invasive nature, real-time imaging, low cost

Introduction

and also there is no risk of exposure to any harmful radiation during the procedure [1,2]. Ultrasound is a sound wave with a frequency greater than 20 kHz. These sound waves are transmitted into the body to capture live images of the

* Corresponding author at: CSIR-Central Scientific Instruments Organisation (CSIR-CSIO), Chandigarh 160030, India. E-mail addresses: [email protected] (Kriti), [email protected] (J. Virmani), [email protected] (R. Agarwal). https://doi.org/10.1016/j.bbe.2018.10.002 0208-5216/© 2018 Nalecz Institute of Biocybernetics and Biomedical Engineering of the Polish Academy of Sciences. Published by Elsevier B.V. All rights reserved. Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

18 19 20 21

BBE 310 1–17

2 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

internal structures of the body like joints, tendons, muscles and internal organs like kidney, liver, etc. For the formation of an ultrasound image, first a sound wave is transmitted from the ultrasound scanner into the human body using a transducer. Inside the body, these sound waves are reflected from different tissues or structures and return to the transducer. The transducer then converts reflected echoes into electrical pulses which are then processed in the scanner to produce the image. The scanner computes the distance of probe from the tissue using velocity of sound and the time of return for each echo. By displaying the distances and echo intensities on the screen, a 2D ultrasound image is formed [3]. The visual quality of ultrasound image is an important factor to effectively diagnose any abnormality present in the internal body structure. For an ultrasound image, the visual quality is hampered by low contrast and presence of speckle noise and this makes the interpretation of the image quite difficult [4,5]. Speckle noise degrades the image quality by masking the detailed structures in the image which are diagnostically important thereby affecting the radiologist's visual interpretation. It is important to remove speckle from the medical images so that the image interpretation as well as accuracy of computer assisted algorithms in the characterization of these images can be improved [1–3,5–7,12,16,17,21]. Different despeckling filter algorithms have been widely used for analysis, segmentation and characterization of medical images [5–26]. A brief summary of the studies carried out by different researchers is given in Table 1. Based on the above literature review it can be concluded that despeckle filtering algorithms yield better segmentation and characterization of different diseases of various body organs. The radiologists visualize the shape of tumour as well as

texture of the region inside the tumour for differential diagnosis between benign and malignant tumours. Therefore, it is expected that a despeckle filtering algorithm which yields better edge and feature/structure preservation would yield optimal results for segmentation as well as characterization of tumours in medical images. In the present work, the assessment of 42 despeckle filtering algorithms has been done for segmentation of breast tumours from ultrasound images.

2.

54 55 56 57 58 59 60 61 62

Methodology adopted

The methodology adopted for assessment of despeckle filtering algorithms for segmentation of breast tumours from ultrasound images is shown in Fig. 1.

63 64 65

2.1.

66

Breast ultrasound image database

Total 100 breast ultrasound images (40 benign and 60 malignant) taken from a standard benchmark database of ultrasound images available at [27] are processed in the present work using 42 despeckle filtering algorithms. A brief description of the database is given in Fig. 2.

67 68 69 70 71

2.2.

72

Despeckling module

Speckle is a multiplicative noise, having a granular structure formed due to superposition of echoes having random phases and amplitudes. The speckle noise ranges from zero to maximum depending upon the constructive or destructive interference of the echoes [3]. The speckle noise can be modelled as:

73 74 75 76 77 78 79

Table 1 – Training course agenda for postcardiac arrest syndrome (PCAS). Start Finish

Contents

9:00 9:15 9:20 9:30 9:35 10:05 10:35

9:15 9:20 9:30 9:35 10:05 10:35 11:05

Reception Introduction Lecture Pause Skill 1 Skill 2 Skill 3

11:05 11:45 11:55 12:45 12:55 13:45 13:55 14:45 14:55 15:10 15:15 16:10

11:45 11:55 12:45 12:55 13:45 13:55 14:45 14:55 15:10 15:15 16:10 16:20

Lecture Pause Skill 4 Pause Skill 5 Pause Skill 6 Pause Debriefing Pause Skill 7 Summary

Course introduction Orientation Break Airway management in PCAS Brain monitoring Neurological physical examination Neurocritical care Break PCPS, IABP Break TTM Break Simulation: VF!PCAS Break Debriefing for simulation training Break Simulation: VF(ECPR) Course conclusion

ECPR, extracorporeal cardiopulmonary resuscitation; IABP, intraaortic balloon pumping; PCPS, percutaneous cardiopulmonary support; TTM, targeted temperature management; VF, ventricular fibrillation.

yði; jÞ ¼ xði; jÞnði; jÞ

(1)

y(i, j) is the noisy image, x(i, j) is the noise free image and n(i, j) is the multiplicative noise. Some despeckling filters work on the multiplicative noise but for some filters the multiplicative noise needs to be converted to additive noise. This transformation is done using log operation.

80 81 82 83 84 85 86

log½yði; jÞ ¼ log½xði; jÞnði; jÞ ¼ log½xði; jÞ þ log½nði; jÞ

(2) 87 88 89

Eq. (2) can be rewritten as:

90 yi;j ¼ xi;j þ ni;j

(3)

Different techniques have been proposed for the reduction of speckle noise [3,9,16,19,28–48]. The despeckle filtering algorithms are broadly classified into eight categories namely local statistics based filters, fuzzy filters, Fourier filters, multiscale filters, non-linear iterative filters, total variation filters, non-local mean filters and hybrid filters as shown in Fig. 3. The parameters used in the implementation of each filter are selected with reference to [3,5,7,9,16,19,28,31,35–48] and are shown in Table 2.

2.2.1.

Local statistics based filters

In these filters, the filtering operation is based on the local statistics. Here, a weighted average is computed

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

93 91 92 94 95 96 97 98 99 100 101 102 103 104

BBE 310 1–17

3

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

Fig. 1 – Methodology adopted for assessment of despeckle filtering algorithms for segmentation of breast tumours from ultrasound images. Note: * Average value of Jaccard index obtained from 104 tumours (43 benign, 61 malignant).

105 106 107 108

using sub-region statistics to calculate the statistical measures over different pixel windows with the window size varying from 3 to 15. The basic form of these filters is given as

(4)

109

^i;j is pixel value of denoised image in the moving window, yi,j is y pixel value of noisy image, yi;j is local mean and wi;j is weighing factor and (i, j) are the pixel co-ordinates. Despeckling filters

110 111 112 113

^i;j ¼ yi;j þ wi;j ðyi;j yi;j Þ y

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

BBE 310 1–17

4

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

Fig. 2 – Description of breast ultrasound image database.

Fig. 3 – Different despeckle filtering algorithms used in the present work. Note: Lsminsc: local statistics minimum speckle index, Lsmv: local statistics mean variance, FBL: fast bilateral, TMED: triangulation median, TMAV: triangulation moving average, ATMED: asymmetrical triangulation median, ATMAV: asymmetrical triangulation moving average, FI: Fourier ideal, FB: Fourier Butterworth, HFI: homomorphic Fourier ideal, HFB: homomorphic Fourier Butterworth, MPT: multiscale product thresholding, IOWT: inter orthonormal wavelet thresholding, NSS: NeighShrinkSure, TV: total variation, ATV: adaptive total variation, HyMedian: hybrid median, Homog: maximum homogeneity, Homo: homomorphic, SRAD: speckle reducing anisotropic diffusion, AD: anisotropic diffusion, DPAD: detail preserving anisotropic diffusion, LS: linear scaling, Ecasort: linear scaling and sorting, OBNLM: optimized Bayesian non-local means, PPB: probabilistic patch based, HyTMED: hybrid triangulation median, HyTMAV: hybrid triangulation moving average, HyATMED: hybrid asymmetrical triangulation median, HyATMAV: hybrid asymmetrical triangulation moving average, GW: geometric Wiener.

114 115 116 117 118 119 120 121 122 123

used under this category are: Lee filter [28], Lee Sigma filter [29,49], Kuan filter [30], Frost filter [31], local statistics minimum speckle index (Lsminsc) filter [3,16], local statistics mean variance (Lsmv) filter [16], Wiener filter [3,4,16] and fast bilateral (FBL) filter [32,48]. The ultrasound images of benign tumour in original image and images despeckled by local statistics based filters are shown in Fig. 4. A blocky effect is observed in the US images despeckled using Lee Sigma filter (Fig. 4(c)), Kuan filter (Fig. 4(d)) and Frost filter (Fig. 4(e)). Smoothing is observed in the US images after

the application of FBL filter (Fig. 4(i)) resulting in loss of information. A slight blurring is also noticed in the images filtered using the Lsmv filter (Fig. 4(g)).

2.2.2.

Fuzzy filters

The fuzzy filters are implemented by applying weighted functions of fuzzy membership-type to the image pixels within a moving window. These filters can be used to remove random noise or impulse noise [14,33,34]. The general form of fuzzy filters is given as

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

124 125 126 127 128 129 130 131 132

BBE 310 1–17 biocybernetics and biomedical engineering xxx (2018) xxx–xxx

5

Table 2 – Parameters used for implementation of despeckling filters. Despeckling filter

Parameters

Local statistics based filter

Lee [5,7,19,24,28] Lee Sigma [29] Kuan [7,19,24,30] Frost [5,7,19,24,31] Lsminsc [3,16,19,24] Lsmv [3,16,19,24] Wiener [3,16,19,24] FBL [19,24,32]

Fuzzy filters

TMED [13,14,19,24,33,34] TMAV [13,14,19,24,33,34] ATMED [13,14,19,24,33,34] ATMAV [13,14,19,24,33,34] FI/HFI [9,19,24] FB/HFB [9,19,24] MPT [19,24,36] IOWT [39] BlockShrink [38] BayesShrink [19,35] NSS [19,24,37] TV [19,24,40] ATV [19,41] Shock [45] Median [3,16,19] HyMedian [3,16,19] Homog [3,16,19] Homo [3,16,19] Geometric [3,7,16,19] SRAD [7,16,19,43] AD [3,7,16,19,42] DPAD [5,7,19,44] CA [3,16] LS [3,16] Ecasort [3,16] OBNLM [5,19,46] PPB [5,19,47] HyTMED [13,14,19,24,34] HyTMAV [13,14,19,24,34] HyATMED [13,14,19,24,34] HyATMAV [13,14,19,24,34] GW [19]

Fourier filters Multiscale filters

Total variation filters Non-linear iterative filters

Non-local mean filters Hybrid filters

133

P ^ði; jÞ ¼ y

Y½yði þ m; j þ nÞyði þ m; j þ nÞ P ðm;nÞ 2 A Y½yði þ m; j þ nÞ

ðm;nÞ 2 A

Neighbourhood size = 5, No. of iterations = 2 Neighbourhood size = 5 No. of iterations = 2 Neighbourhood size = 5 Neighbourhood size = 5, No. of iterations = 2, Edge = 0 Neighbourhood size = 5, No. of iterations = 2 Neighbourhood = 5, noise = [] Width of spatial Gaussian (sigmaS) = 10, Width of range Gaussian (sigmaR) = 20 Window size = 3  3 Window size = 3  3 Window size = 3  3 Window size = 3  3 Cut off frequency = 500 Cut off frequency = 500, Order = 2 Noise variance (v) = 28, Scale number = 2, c = 12 Wavelet type = sym8 Low frequency cut off for Shrinkage Jmin = 5, wavelet filter sym8 Wavelet type = haar, levels = 2 Decomposition level L = 3, wavelet type = sym8 u = 15, No. of iterations = 5, dt = 0.25, g = 1 m = 20, tol = 0.2, l = 1 Mask size = 9, No. of iterations = 30, dt = 0.25 Neighbourhood size = 5, No. of iterations = 2 Neighbourhood size = 5, No. of iterations = 3 Neighbourhood size = 5, No. of iterations = 3 Neighbourhood size = 5 Neighbourhood size = 3, No. of iterations = 2 Time step = 0.02, No. of steps = 30 No. of iterations = 20, kappa = 30, lambda = 0.25, option = 1 Step size = 0.02, No. of iterations = 30, Cu noise estimation with n = 5 Neighbourhood size = 5, No. of iterations = 2 Neighbourhood size = 5, No. of iterations = 3 Neighbourhood size = 5, No. of iterations = 3 M = 11, alpha = 7, smoothing parameter h = 0.4, offset = 100 hW = 12, hD = 4, a = 0.8, T = 2, nbit = 4, L = 1 Window size for fuzzy and Wiener filters = 3  3 Window size for geometric filter = 3  3 and No. of iterations = 2

(5)

134 135 136

Y[y(i, j)] represents the window function, A is the area, y(i, j) is the noisy image.

137 138 139 140 141 142 143 144 145 146

Different filters used under this category are: triangulation median (TMED) filter, triangulation moving average (TMAV) filter, asymmetrical triangulation median (ATMED) filter and asymmetrical triangulation moving average (ATMAV) filter. The ultrasound images of benign tumour in original image and images despeckled by fuzzy filters are shown in Fig. 5. It is observed that a blocky effect as well as blurring is introduced in the images after the application of ATMAV filter (Fig. 5(e)).

147 148 149

2.2.3.

Fourier filters

These filters are based on the Fourier transform. First, the image is converted to the transform domain using Fourier

transform and then the image is filtered using either ideal or Butterworth filters followed by inverse Fourier transform. The same procedure is followed at the time of homomorphic filters except before transforming the image from spatial to transform domain, the image is first projected onto a logarithmic space [9]. The filters that come under this category are: Fourier ideal (FI) filter, Fourier Butterworth (FB) filter, homomorphic Fourier ideal (HFI) filter and homomorphic Fourier Butterworth (HFB) filter. The ultrasound images of benign tumour in original image and images despeckled by Fourier filters are shown in Fig. 6.

2.2.4.

Multiscale filters

These noise reduction techniques are also known as thresholding or wavelet shrinkage techniques. These techniques are based on the principle of transforming the original image into different scales. The different filters that come under this category are: BayesShrink filter [35], multiscale product thresholding (MPT) filter [36], NeighShrinkSure (NSS) filter

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167

BBE 310 1–17

6

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

Fig. 4 – Ultrasound images of benign tumour (a) Original image, Images despeckled by (b) Lee filter, (c) Lee Sigma filter, (d) Kuan filter, (e) Frost filter, (f) Lsminsc filter, (g) Lsmv filter, (h) Wiener filter, (i) FBL filter.

Fig. 5 – Ultrasound images of benign tumour (a) Original image, Images despeckled by (b) TMED filter, (c) TMAV filter, (d) ATMED filter, (e) ATMAV filter.

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

BBE 310 1–17 biocybernetics and biomedical engineering xxx (2018) xxx–xxx

7

Fig. 6 – Ultrasound images of benign tumour (a) Original image, Images despeckled by (b) FI filter, (c) FB filter, (d) HFI filter, (e) HFB filter.

Fig. 7 – Ultrasound images of benign tumour (a) Original image, Images despeckled by (b) MPT filter, (c) IOWT filter, (d) BlockShrink filter, (e) BayesShrink filter, (f) NSS filter.

168 169 170 171

[37], BlockShrink filter [38] and inter orthonormal wavelet thresholding (IOWT) filter [39]. The ultrasound images of benign tumour in original image and images despeckled by multiscale filters are shown in Fig. 7.

172 173 174

2.2.5.

Total variation filters

Total variation denoising was developed to reduce the noise in an image while preserving the sharp edges. The basic concept

of these filters was coined by Rudin, Osher and Fatemi in 1992 [40]. The denoising by total variation can be viewed as an optimization problem wherein the filter output can be obtained by minimizing a particular cost function. Different filters used under this category are: total variation (TV or ROF) filter [40] and anisotropic total variation (ATV) filter [41]. The ultrasound images of benign tumour in original image and images despeckled by total variation filters are shown in Fig. 8.

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

175 176 177 178 179 180 181 182

BBE 310 1–17

8

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

Fig. 8 – Ultrasound images of benign tumour (a) Original image, Images despeckled by (b) TV filter, (c) ATV filter.

Fig. 9 – Ultrasound images of benign tumour (a) Original image, Images despeckled by (b) Shock filter, (c) Median filter, (d) HyMedian filter, (e) Homog filter, (f) Homo filter, (g) Geometric filter, (h) SRAD filter, (i) AD filter, (j) DPAD filter, (k) CA filter, (l) LS filter, (m) Ecasort filter.

183

2.2.6. 184 185 186 187 188 189

Non-linear iterative filters

The filters that come under this category are: Geometric filter [3,16], maximum homogeneity (Homog) filter [9], homomorphic (Homo) filter, anisotropic diffusion (AD) filter [42], speckle reducing anisotropic diffusion (SRAD) filter [43], median filter [3,16], hybrid median (HyMedian) filter [3,16], detail preserving anisotropic diffusion (DPAD) filter [44], Linear scaling (LS) filter

[3,16], CA filter [3,16], linear scaling and sorting (Ecasort) filter [3,16] and shock filter [45]. The ultrasound images of benign tumour in original image and images despeckled by non-linear iterative filters are shown in Fig. 9. Blurring is observed in the images filtered using median filter (Fig. 9(c)), AD filter (Fig. 9(i)) CA filter (Fig. 9(k)), LS filter (Fig. 9(l)) and Ecasort filter (Fig. 9(m)) that results in loss of texture information.

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

190 191 192 193 194 195 196

BBE 310 1–17 biocybernetics and biomedical engineering xxx (2018) xxx–xxx

9

Fig. 10 – Ultrasound images of benign tumour (a) Original image, Images despeckled by (b) OBNLM filter, (c) PPB filter.

Fig. 11 – Ultrasound images of benign tumour (a) Original image, Images despeckled by (b) HyTMED filter, (c) HyTMAV filter, (d) HyATMED filter, (e) HyATMAV filter, (f) GW filter.

197

2.2.7.

198 199 200 201 202 203 204 205 206 207 208

These filters work on the concept of patches around each pixel in an image that are used to create the metric for the filters' operation. The performance parameters of these filters are the size of search window, similarity window and the value of smoothing parameter h. The filters considered under this category are: optimized Bayesian non-local means (OBNLM) based filter [46] and probabilistic patch based (PPB) filter [47]. The ultrasound images of benign tumour in original image and images despeckled by non-local mean filters are shown in Fig. 10. It is observed that the images filtered using PPB filter (Fig. 10(c)) show a smoothening effect that results in loss of information.

209 210 211 212 213 214 215

2.2.8.

Non-local mean filters

Hybrid filters

Different hybrid filters are formed by the combination of fuzzy filters, geometric filter with wiener filter [17,33]. The filters used under this category are: hybrid triangulation median (HyTMED) filter, hybrid triangulation moving average (HyTMAV) filter, hybrid asymmetrical triangulation median (HyATMED) filter, hybrid asymmetrical triangulation moving

average (HyATMAV) filter and geometric wiener (GW) filter. The ultrasound images of benign tumour in original image and images despeckled by hybrid filters are shown in Fig. 11. A blocky effect is observed in images after the application of HyATMAV filter (Fig. 11(e)).

216 217 218 219 220 221

Assessment of despeckle filtering algorithms

222

The assessment of despeckle filtering algorithms has been done both on the basis of objective assessment as well as subjective assessment. The objective assessment has been done on the basis of edge and feature/structure preservation capabilities using image quality metrics and the subjective assessment is carried out by participating experienced radiologist.

223 224 225 226 227 228 229

2.3.

2.3.1.

Objective assessment of despeckle filtering algorithms

The exhaustive review of literature [3,5,8,13–20,22–24] indicates that a number of metrics/measures, i.e. image quality

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

230 231 232

BBE 310 1–17

10

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

Table 3 – Image quality metrics used for objective assessment of despeckle filtering algorithms. Metric name

Significance Used to find the level of speckle before and after the filtering process. Finds the objective difference between original and despeckled images. Used to estimate the mean difference between original and despeckled images. It is the square root of squared average error over a window. Used to find the edge features of the image. Used to find the degree of closeness between original and despeckled images. Average of the difference between original and despeckled images. Maximum value of error between original and despeckled images. Used to find the similarity between original and despeckled images. Used to measure the error prediction accuracy. Norm of dissimilarity between original and despeckled images. Measurement of alignment before and after despeckling. Measures similarity between the noisy and denoised images. Measures the edge preservation capability of the despeckling filter. Used to model the distortion.

Signal-to-noise ratio (SNR) [3,13,14,16,19,25] Peak signal-to-noise ratio (PSNR) [3,5,7,13,14,16,19] Mean square error (MSE) [3,5,7,13,14,16,19] Root mean square error (RMSE) [3,16] Laplacian mean square error (LMSE) [3,16,19] Correlation coefficient (CC) [3,16,19]

Q3

233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260

Average difference (AD) [3,16] Maximum difference (MD) [3,16,19] Structural content (SC) [3,16,19] Normalized average error (NAE) [3,16,19] Normalized error summation (Err3, Err4) [3,16,19] Normalized cross-correlation (NK) [3,16,19] Structural similarity index (SSIM) [3,14,16,19,50] Beta metric (b) [1,9,14,17,19] Image quality index (IQI) [3,16,19,24,51]

metrics have been used for the objective assessment of despeckle filtering algorithms on medical images. A brief summary of these image quality metrics is described in Table 3. For the objective assessment of despeckle filtering algorithms edge, feature/structure preservation has been considered. An ideal despeckling algorithm should retain edges while smoothing the homogeneous areas thereby preserving features/structures of the image. It has been observed from the exhaustive literature review that for the quantification of edge preservation capabilities of the despeckling algorithm, beta metric (b) is used [1,9,14,19,24] while feature/structure preservation can be adequately quantified by evaluating IQI metric [3,16,19,24,51]. The traditional parameters for image quality evaluation like MSE, PSNR, MD, SNR, etc., also sometimes known as full reference image quality metrics, often fail to show the true performance of the despeckling filters due to the absence of a noise free reference input image [5,24] therefore the true performance of despeckling filters is evaluated on the basis of edge preservation and feature/structure preservation capabilities. For the differential diagnosis between benign and malignant breast ultrasound tumours, both the texture and shape of the tumour are important therefore an ideal despeckle filtering algorithm would be the one that yields higher values of both b and IQI. In the present study a despeckle filtering algorithm that yields the highest average value of b and IQI metric is considered optimal. The b and IQI metrics are given as:

261 IQI ¼ 262 263 264 265 266 267

s od 2o ( d ( 2s o s d : : s o s d ðo ( Þ2 þ ðd ( Þ2 s 2o þ s 2d

(6)

sod: covariance between original and despeckled image, so: standard deviation of original image, sd: standard deviation of despeckled image, ō: mean of original image, d: mean of despeckled image.   P  R;C DIo DIo DId DId (7) b¼P  2  2 DId DId R;C DIo DIo

DIo is the high pass filtered version of Io: original image, DIo: mean of DIo, DId: High pass filtered version of Id despeckled image, DId : mean of DId.

268 269 270 271 272

b þ IQI Average value ¼ 2

(8)

These image quality metrics have been obtained using the original image and the corresponding despeckled image for all the 42 despeckle filtering algorithms and the results are represented in the form of mean (m)  standard deviation (SD). (a) Objective assessment of local statistics based filters The breast ultrasound images are filtered using eight local statistics based filters yielding 100 pre-processed images for each filter. The performance of each filter in terms of average value of b and IQI for these pre-processed images is shown in Table 4. It is observed from Table 4 that the filters Lee Sigma and Wiener show good edge preservation capabilities with b > 0.90 while Lee sigma shows highest edge preservation with b value 0.98  0.04. The IQI value for each of these filters is above 0.95. However, it is observed that among all the local statistics based filters, highest average value of b and IQI is obtained for the Lee Sigma filter. (b) Objective assessment of fuzzy filters The breast ultrasound images are filtered using four fuzzy filters yielding 100 pre-processed images for each filter. The performance of each filter in terms of average value of b and IQI for these pre-processed images is shown in Table 5. It is observed from Table 5 that ATMED filter results in highest edge preservation with b value 0.90  0.04. The IQI value for each of these filters is above 0.95. However, it is observed that among all the fuzzy filters, highest average value of b and IQI is obtained for ATMED filter. (c) Objective assessment of Fourier filters The breast ultrasound images are filtered using four Fourier filters yielding 100 pre-processed images for each

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307

BBE 310 1–17

11

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

Table 4 – Image quality metrics (m W SD) computed for breast ultrasound images filtered using local statistics based filters. Quality metrics b IQI Average value

Lee

Lee Sigma

Kuan

Frost

Lsminsc

Lsmv

Wiener

FBL

0.79  0.07 0.99  0.00 0.89  0.11

0.98  0.04 0.98  0.04 0.98 W 0.04

0.44  0.19 0.98  0.04 0.71  0.30

0.73  0.10 0.98  0.04 0.86  0.14

0.52  0.09 0.99  0.01 0.76  0.24

0.78  0.07 0.99  0.00 0.89  0.11

0.92  0.02 0.99  0.00 0.96  0.04

0.84  0.05 0.97  0.02 0.91  0.07

Table 5 – Image quality metrics (m W SD) computed for breast ultrasound images filtered using fuzzy filters. Quality metric b IQI Average value

TMED

TMAV

ATMED

ATMAV

0.54  0.26 0.99  0.00 0.77  0.29

0.76  0.15 0.99  0.00 0.88  0.16

0.90  0.04 0.99  0.01 0.95 W 0.05

0.58  0.18 0.98  0.01 0.78  0.24

Table 6 – Image quality metrics (m W SD) computed for breast ultrasound images filtered using Fourier filters. Quality metric b IQI Average value

FI

FB

HFI

HFB

0.99  0.00 0.99  0.00 0.99 W 0.00

0.98  0.00 1.00  0.00 0.99 W 0.00

0.96  0.05 0.99  0.00 0.97  0.06

0.99  0.00 0.99  0.00 0.99 W 0.00

Table 7 – Image quality metrics (m W SD) computed for breast ultrasound images filtered using multiscale filters. Quality Metric b IQI Average value

310 311 312 313 314 315 316 317 318 320 319 321 322 323 324 325 326 327 328 329 330 331 333 332 334 335 336 337 338 339 340 341

MPT

IOWT

BlockShrink

BayesShrink

NSS

0.97  0.00 0.99  0.00 0.98  0.00

0.98  0.00 0.98  0.00 0.98  0.00

0.98  0.00 0.99  0.00 0.98  0.00

0.99  0.00 1.00  0.00 0.99 W 0.00

0.97  0.00 0.98  0.00 0.97  0.00

filter. The performance of each filter in terms of average value of b and IQI for these pre-processed images is shown in Table 6. It is observed from Table 6 that all the Fourier filters show good edge preservation capabilities with b ≥ 0.95. The IQI value for each of these filters is above 0.95. However, it is observed that among all the Fourier filters, highest average value of b and IQI is obtained for FI, FB and HFB filters. (d) Objective assessment of multiscale filters The breast ultrasound images are filtered using five multiscale filters yielding 100 pre-processed images for each filter. The performance of each filter in terms of average value of b and IQI for these pre-processed images is shown in Table 7. It is observed from Table 7 that all the multiscale filters result in good edge preservation with b value above 0.95. The IQI value for each of these filters is also above 0.95. However, it is observed that among all the multiscale filters, highest average value of b and IQI is obtained for BayesShrink filter. (e) Objective assessment of total variation filters The breast ultrasound images are filtered using two total variation filters yielding 100 pre-processed images for each filter. The performance of each filter in terms of average value of b and IQI for these pre-processed images is shown in Table 8. It is observed from Table 8 that both TV and ATV filters exhibit similar performance with almost same average value of b and IQI.

(f) Objective assessment of non-linear iterative filters The breast ultrasound images are filtered using twelve non-linear iterative filters yielding 100 pre-processed images for each filter. The performance of each filter in terms of average value of b and IQI for these pre-processed images is shown in Table 9. It is observed from Table 9 that all the filters other than SRAD, DPAD and HyMedian result in poor edge preservation with b ≤ 0.75. SRAD and HyMedian filters have moderate edge preservation capabilities while DPAD filter results in highest edge preservation with b ≥ 0.95. The IQI value for each of these filters is above 0.95. However, it is observed that among all the non-linear iterative filters, highest average value of b and IQI is obtained for the DPAD filter. (g) Objective assessment of non-local mean filters The breast ultrasound images are filtered using two non-local mean filters yielding 100 pre-processed images for each filter. The performance of each filter in terms of average value of b and IQI for these pre-processed images is shown in Table 10. Table 8 – Image quality metrics (m W SD) computed for breast ultrasound images filtered using total variation filters. Quality metric b IQI Average value

TV

ATV

0.84  0.05 0.99  0.01 0.92 W 0.08

0.83  0.06 0.99  0.00 0.91  0.09

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 359 358 360 361 362 363 364 365 366 367 368 369

BBE 310 1–17

12

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

Table 9 – Image quality metrics (m W SD) computed for breast ultrasound images filtered using non-linear iterative filters. Quality metric b IQI Average value

Quality metric b IQI Average value

Shock

Median

HyMedian

Homog

Homo

Geometric

0.46  0.05 0.99  0.01 0.72  0.26

0.70  0.09 0.99  0.01 0.84  0.15

0.90  0.03 0.99  0.00 0.95  0.05

0.62  0.10 0.99  0.03 0.80  0.19

0.72  0.08 0.99  0.01 0.85  0.14

0.70  0.15 0.98  0.02 0.84  0.17

SRAD

AD

DPAD

CA

LS

Ecasort

0.89  0.05 0.99  0.00 0.94  0.06

0.22  0.03 0.98  0.02 0.60  0.38

0.99  0.00 0.99  0.00 0.99 W 0.00

0.71  0.12 0.99  0.01 0.85  0.16

0.32  0.14 0.98  0.01 0.65  0.34

0.71  0.12 0.99  0.01 0.85  0.16

Table 10 – Image quality metrics (m W SD) computed for breast ultrasound images filtered using non-local mean filters. Quality metric b IQI Average value

OBNLM

PPB

0.90  0.02 0.94  0.05 0.92 W 0.04

0.71  0.10 0.98  0.01 0.85  0.15

365 366 367 368 369

It is observed from Table 10 that OBNLM filter results in better edge preservation than the PPB filter. The IQI value for both these filters is above 0.90. However, it is observed that among all the non-local mean filters, highest average value of b and IQI is obtained for the OBNLM filter.

370 371 372 373 374 375 376 377 378 379 380 381 382

(h) Objective assessment of hybrid filters The breast ultrasound images are filtered using five hybrid filters yielding 100 pre-processed images for each filter. The performance of each filter in terms of average value of b and IQI for these pre-processed images is shown in Table 11. It is observed from Table 11 that HyATMED filter results in best edge preservation as compared to other filters with b value 0.84  0.05. The IQI value for each of these filters is above 0.95. However, among all the hybrid filters, highest average value of b and IQI is obtained for HyATMED filter.

383 384 385 386 387 388 389 390 391 392 393 394

2.3.1.1. Optimal despeckle filtering algorithms based on objective assessment. Based on the objective assessment of all the 42 despeckle filtering algorithms, comparative analysis of the filters yielding highest average value of b and IQI from each filter category is shown in Table 12. Out of the despeckle filtering algorithms selected from each filter category as shown in Table 12, filters with high average value of b and IQI (>0.95) are considered optimal for edge, feature/structure preservation. The despeckled images obtained from these optimal filters are used for further assessment. The optimal despeckle filtering algorithms are Lee Sigma, FI, FB, HFB, BayesShrink and DPAD.

2.3.2. Subjective assessment of despeckle filtering algorithms (clinical validation by radiologist)

395 396

The 100 pre-processed breast ultrasound images obtained from the filters labelled as optimal based on the highest average value of b and IQI are only considered for the subjective assessment. These images are shown to an experienced radiologist on an LCD screen under uniform lighting conditions. The protocol followed by the radiologist for grading the images are (a) Tumour delineation: The tumour should be clearly distinguishable in the despeckled image, (b) Boundary of the tumour: The boundary/shape characteristics of the tumour should be clearer after despeckling, (c) Texture of the region inside tumour: The texture of the region inside tumour should be same after despeckling as in original image, i.e. no blocking or echoes should be introduced after despeckling, (d) Blurring in the image: There should be no blurring of the image as it reduces the visual quality and (e) Sonographic appearance of the breast tissue surrounding the tumour: The appearance of the surrounding breast tissue should be nearly same after despeckling as in original image. Based on the visual assessment of the despeckled image in comparison to the original image in accordance with the above protocols, a grade is assigned to each despeckled image from 1 to 5 (here 1 denotes poor image quality and 5 denotes high image quality).

397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420

2.3.2.1. Optimal despeckle filtering algorithms based on subjective assessment. Based on the grades assigned by the

421 422 423 424 425 426 427 428 429 430 431 432 433

radiologist, the final grade is obtained as the average of the grades given to all the 100 (40 benign and 60 malignant) preprocessed images. The final average grade obtained for each filter is tabulated in Table 13. It is observed from Table 13 that despeckled images obtained from the filters BayesShrink and DPAD are clinically acceptable while the images obtained from the Lee Sigma filter are not acceptable in clinical practice. From the objective and subjective assessment carried out for the despeckle filtering algorithms it has been observed that on the basis of objective assessment, six despeckle filtering

Table 11 – Image quality metrics (m W SD) computed for breast ultrasound images filtered using hybrid filters. Quality metric

HyTMED

HyTMAV

HyATMED

HyATMAV

GW

b IQI Average value

0.62  0.27 0.99  0.00 0.81  0.26

0.75  0.16 0.99  0.00 0.87  0.16

0.84  0.05 0.99  0.00 0.92 W 0.08

0.29  0.11 0.98  0.06 0.64  0.35

0.70  0.17 0.98  0.01 0.84  0.18

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

BBE 310 1–17

13

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

Table 12 – Comparative analysis of despeckle filtering algorithms yielding maximum average value of b and IQI from each filter category.

Table 14 – Average value of Jaccard Index obtained from contours marked in original images and tumour contours obtained by segmentation process in despeckled images.

Filter category

Image

Local statistics based filters Fuzzy filters Fourier filters

Multiscale filters Total variation filters Non-linear iterative filters Non-local mean filters Hybrid filters

434 435 436 437 438 439

Filter Lee Sigma ATMED FI FB HFB BayesShrink TV DPAD OBNLM HyATMED

Average value of b and IQI 0.98  0.04 0.95  0.05 0.99  0.00 0.99  0.00 0.99  0.00 0.99  0.00 0.92  0.08 0.99  0.00 0.92  0.04 0.92  0.08

Jaccard Index* **

Original images Despeckled images Despeckled images Despeckled images Despeckled images Despeckled images Despeckled images

using using using using using using

Lee Sigma filter FI filter FB filter HFB filter BayesShrink filter DPAD filter

79.01% 79.48% 79.33% 79.19% 78.99% 79.46% 79.52%

* Average value of Jaccard index obtained from 104 tumours (43 benign, 61 malignant). ** Segmentation results obtained using tumour contours marked by radiologist in original images and tumour contours obtained by segmentation process in original images.

Table 13 – Grading obtained for optimal despeckle filtering algorithms.

2.4.

Filter name

Grade assigned

Lee Sigma FI FB HFB BayesShrink DPAD

1 3 3.5 3 4 4.5

Segmentation is the process of extracting the desired region of interest (ROI) from an image using some automatic or semiautomatic process. In case of breast ultrasound images, segmentation techniques are used to alienate the tumours from the background [52]. Several algorithms have been developed by many researchers for the segmentation of medical images [52–57]. Out of all the segmentation techniques, active contour method of segmentation has been widely used in case of medical images [57–74]. The active contour models can be classified to be either edge-based or regionbased. The most common example of edge-based active contour models is geodesic active contour model, in which the image gradient information is used to guide the motion of the curve for detecting the target in a background [75]. The example for region-based model is the active contour model without edges as proposed by Chan and Vese [76]. This model

algorithms are considered to be optimal while on the basis of subjective assessment, only two despeckle filtering algorithms are labelled to be best. Therefore, for further assessment the pre-processed images obtained from the optimal despeckle filtering algorithms based on the objective assessment are fed to the segmentation module.

Segmentation module: active contour model

Fig. 12 – Segmentation of tumour from breast ultrasound images (a) Original image indicating tumour boundary marked by the radiologist, (b) Original image indicating tumour boundary obtained after segmentation, (c) Despeckled image using Lee Sigma filter indicating tumour boundary obtained after segmentation, (d) Despeckled image using FI filter indicating tumour boundary obtained after segmentation, (e) Despeckled image using FB filter indicating tumour boundary obtained after segmentation, (f) Despeckled image using HFB filter indicating tumour boundary obtained after segmentation, (g) Despeckled image using BayesShrink filter indicating tumour boundary obtained after segmentation, (h) Despeckled image using DPAD filter indicating tumour boundary obtained after segmentation.

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456

BBE 310 1–17

14

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

Fig. 13 – The sample breast ultrasound images pre-processed using best performing DPAD filter followed by segmentation: (a) benign image; (b) malignant image.

457 458 459 460 461 462 463 464 465

of segmentation is used to detect the object boundaries in a region having a homogeneous intensity distribution. In the present work, the Chan and Vese model has been implemented to segment the breast tumours from ultrasound images. The true boundary of 104 breast tumours (43 benign, 61 malignant) from 100 breast ultrasound images have been delineated in the presence of experienced radiologist using Image J software [77]. The sample images showing the results of segmentation are shown in Fig. 12.

466 467 468 469 470 471

2.4.1.

Objective assessment of segmentation algorithm

The efficacy of the segmentation algorithm is calculated using the area overlap between the original tumour contour marked by a radiologist and the tumour contour obtained as a result of segmentation. The area overlap is calculated using Jaccard index (also referred to as overlap or similarity) [72–75] given as:

472 Jaccard index ¼

AG \ AS 100 AG [ AS

(9)

473 474 475

AG: area of the ground truth image (marked by radiologist); AS: area of the segmented image

476 477 478 479 480 481 482

The segmentation results in terms of Jaccard index obtained from tumour contours marked in original images and tumour contours obtained by segmentation process in despeckled images are shown in Table 14. The results in Table 14 show that the segmentation performance has degraded in case of tumours extracted from the images pre-processed using HFB filter while FI, FB filters

show marginal improvement and the optimal filter resulting in best segmentation performance is DPAD filter. Therefore, both quantitative and qualitative assessment of despeckle filtering algorithms for segmentation of breast tumours from ultrasound images indicate that DPAD is the best performing filter. The sample benign and malignant breast ultrasound images pre-processed using best performing DPAD filter followed by segmentation are shown in Fig. 13.

483 484 485 486 487 488 489 490 491

3.

492

Conclusion

In the present work, the assessment of despeckle filtering algorithms for segmentation of breast tumours from ultrasound images has been carried out. The assessment of despeckle filtering algorithms is done quantitatively by obtaining edge and feature/structure preservation capabilities of the filter and qualitatively on the basis of grades provided by the experienced radiologist. The objective assessment of the segmentation algorithm has been carried out using Jaccard index after obtaining the true tumour contours as marked by the radiologist on original images and the tumour contours obtained by segmentation process on the despeckled images. Based on the results obtained by exhaustive experimentation carried out in the present work it has been concluded that the breast ultrasound images pre-processed by DPAD filter are more clinically acceptable and result in better segmentation of benign and malignant tumours. Therefore the study recom-

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508

BBE 310 1–17 biocybernetics and biomedical engineering xxx (2018) xxx–xxx

15

Fig. 14 – Model proposed for segmentation of breast tumours from ultrasound images.

509 510 511 512 513 514

mends the model based on DPAD filter as shown in Fig. 14 for segmentation of tumours from breast ultrasound images. The promising results of the study indicate that the proposed model can be used in routine clinical practice for analysis and segmentation of benign and malignant breast ultrasound images.

515

Conflicts of interest

516

None.

517

Credit author statement

518 519 520 521 522 523 524

Kriti: Conceptualization, methodology, software, validation, investigation, data curation, writing-original draft. Dr. Jitendra Virmani: Conceptualization, methodology, data curation, writing-review and editing, visualization, supervision, project administration. Dr. Ravinder Agarwal: Conceptualization, writing-review and editing, visualization, supervision, resources.

525 526 527

Acknowledgments

528 529 530 531 532 533 534

The authors would like to thank Dr. Shruti Thakur, Kamla Nehru Hospital, Shimla for grading the filtered images and marking the tumour contours in the original ultrasound images. The authors would also like to thank Director, Thapar Institute of Engineering and Technology, Patiala and Director, CSIR-CSIO, Chandigarh for constant patronage and support in carrying out the present research.

references

535

[1] Michailovich OV, Tannenbaum A. Despeckling of medical ultrasound images. IEEE Trans Ultrasonics Ferroelectrics Freq Contr 2006;53(1):64–78. [2] Narayanan SK, Wahidabanu RSD. A view of despeckling in ultrasound imaging. Int J Signal Process Image Process Pattern Recogn 2009;2(3):85–98. [3] Loizou CP, Pattichis CS. Despeckle filtering algorithms and software for ultrasound imaging. Synthesis lectures on algorithms and software for engineering. San Rafael, CA, USA: Claypool Publishers; 2008. [4] Burckhardt CB. Speckle in ultrasound B-mode scans. IEEE Trans Sonics Ultrasonics SU-25 1978;1:1–6. [5] Zhang J, Wang C, Chang Y. Comparison of despeckled filters for breast ultrasound images. Circuits Syst Signal Process 2015;34(1):185–208. [6] Gupta D, Anand RS, Tyagi B. Despeckling of ultrasound medical images using ripplet domain and nonlinear filtering. Signal Image Video Process 2015;9(5):1093–111. [7] Finn S, Glavin M, Jones E. Echocardiographic speckle reduction comparison. IEEE Trans Ultrasonics Ferroelectrics Freq Contr 2011;58(1):82–101. [8] Gupta D, Anand RS, Tyagi B. Despeckling of ultrasound images of bone fracture using M-band ridgelet transform. Optik 2014;125:1417–22. [9] Mateo JL, Fernandez-Caballero A. Finding out general tendencies in speckle noise reduction in ultrasound images. Expert Syst Appl 2009;36(4):7786–97. [10] Wong A, Mishra A, Bizheva K, Clausi DA. General Bayesian estimation for speckle noise reduction in optical coherence tomography retinal imagery. Opt Exp 2010;18(8):8338–52. [11] Elamvazuthi I, Muhd-Zin MLB, Begam KM. Despeckling of ultrasound images of bone fracture using multiple filtering, algorithms. Math Comp Model 2013;57(1–2):152–68. [12] Kocer HE, Cevik KK, Sivri M, Koplay M. Measuring the effect of filters on segmentation of developmental dysplasia of the hip. Iran J Radiol 2016;13(3):e25491.

536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

BBE 310 1–17

16 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640

biocybernetics and biomedical engineering xxx (2018) xxx–xxx

[13] Biradar N, Dewal ML, Rohit MK. A novel hybrid homomorphic fuzzy filter for speckle noise reduction. Biomed Eng Lett 2014;4(2):176–85. [14] Biradar N, Dewal ML, Rohit MK. Edge preserved speckle noise reduction using integrated fuzzy filters. Int Scholarly Res Notices 2014;2014:1–11. [15] Gupta D, Anand RS, Tyagi B. Ripplet domain non-linear filtering for speckle reduction in ultrasound medical images. Biomed Signal Process Contr 2014;10(1):79–91. [16] Loizou CP, Theofanous C, Pantziaris M, Kasparis T. Despeckle filtering software toolbox for ultrasound imaging of common carotid artery. Comp Methods Programs Biomed 2014;14(1):109–24. [17] Loizou CP, Pattichis CS, Christodoulou CI, Istepanian RSH, Pantziaris M, Nicolaides A. Comparative evaluation of despeckle filtering in ultrasound imaging of the carotid artery. IEEE Trans Ultrasonics Ferroelectrics Freq Contr 2005;52(10):1653–69. [18] Gupta D, Anand RS, Tyagi B. Speckle filtering of ultrasound images using a modified non-linear diffusion model in nonsubsampled Shearlet domain. IET Image Process 2015;9 (2):107–17. [19] Biradar N, Dewal ML, Rohit MK. Speckle noise reduction in B-mode echocardiographic images: a comparison. IETE Techn Rev 2015. doi: 10.1080/02564602. 2015.1031714. [20] Manth N, Virmani J, Kumar V, Kalra N, Khandelwal N. Despeckle filtering: performance evaluation for malignant focal hepatic lesions. Proceedings of the 2nd International Conference on Computing for Sustainable Global Development; 2015. [21] Subramanya MB, Kumar V, Mukherjee SD, Saini M. SVMbased CAC system for B-mode kidney ultrasound images. J Digital Imaging 2015;28:448–58. [22] Hiremath PS, Akkasaligar PT, Badiger S. Visual enhancement of digital ultrasound images using multiscale wavelet domain. Pattern Recogn Image Anal 2010;20(3):303–15. [23] Hiremath PS, Akkasaligar PT, Badiger S. Speckle reducing contourlet transform for medical ultrasound images. Int J Comp Elect Automation Contr Inf Eng 2011;5(8):932–9. [24] Biradar N, Dewal ML, Rohit MK. Blind source parameters for performance evaluation of despeckling filters. Int J Biomed Imaging 2016;2016:1–12. [25] Gong G, Zhang H, Yao M. Speckle noise reduction algorithm with total variation regularization in optical coherence tomography. Opt Exp 2015;23(19):24699–712. [26] Stankiewicz A, Marciniak T, Dabrowski A, Stopa M, Rakowicz P, Marciniak E. Denoisisng methods for improving automatic segmentation in OCT images of human eye. Bull Polish Acad Sci Techn Sci 2017;65(1):71–8. [27] http://ultrasoundcases.info/category.aspx?cat=67. [Accessed December 2016]. [28] Lee JS. Speckle analysis and smoothing synthetic aperture radar images. Comp Graphics Image Process 1981;17(1):24–32. [29] Lee JS. Digital image smoothing and sigma filter. Comp Vision Graphics Image Process 1983;24(2):255–69. [30] Kuan DT, Sawchuk AA, Strand TC, Chavel P. Adaptive noise smoothing filter for images with signal-dependent noise. IEEE Trans Pattern Anal Mach Intel 1985;7(2):165–77. [31] Frost VS, Stiles JA, Shanmugan KS, Holtzman JC. A model for radar images and its application to adaptive digital filtering of multiplicative noise. IEEE Trans Pattern Anal Mach Intell 1982;4(2):157–66. [32] Chaudhary KN, Sage D, Unser M. Fast O(1) bilateral filtering using trigonometric range kernels. IEEE Trans Image Process 2011;20(12):3376–82. [33] Kwan HK. Fuzzy filters for noise reduction in images. In: Nachtegael M, Van der Weken D, Kerre EE, Van De Ville D, editors. Fuzzy filters for image processing. Berlin: SpringerVerlag; 2003. p. 25–53.

[34] Biradar N, Dewal ML, Rohit MK. Speckle noise reduction using hybrid TMAV based fuzzy filter. Int J Res Eng Technol 2014;3(3):113–8. [35] Chang SG, Yu B, Vettereli M. Adaptive wavelet thresholding for image denoising and compression. IEEE Trans Image Process 2000;9(9):1532–46. [36] Bao P, Zhang L. Noise reduction for magnetic resonance images via adaptive multiscale products thresholding. IEEE Trans Med Imag 2003;22(9):1089–99. [37] Zhou D, Cheng W. Image denoising with an optimal threshold and neighbouring window. Pattern Recogn Lett 2008;29(11):1694–7. [38] Zhou D, Shen X. Image denoising using block thresholding. Proceedings of 2008 Congress on Image and Signal Processing; 2008. [39] Luisier F, Blu T. A new SURE approach to image denoising: interscale orthonormal wavelet thresholding. IEEE Trans Image Process 2007;16(3):593–606. [40] Rudin LI, Osher S, Fatemi E. Nonlinear total variation based noise removal algorithms. Physica D 1992;60-(1):259–68. [41] Goldstein T, Osher S. The split Bregman method for L1regularized problems. SIAM J Imaging Sci 2009;2(2):423–43. [42] Perona P, Malik J. Scale-space and edge detection using anisotropic diffusion. IEEE Trans Pattern Anal Mach Intell 1990;12(7):629–39. [43] Yu Y, Acton ST. Speckle reducing anisotropic diffusion. IEEE Trans Image Process 2002;11(11):1260–70. [44] Aja-Fernández S, AlberolaLópez C. On the estimation of the coefficient of variation for anisotropic diffusion speckle filtering. IEEE Trans Image Process 2006;15(9):2694–701. [45] Weickert J. Coherence-enhancing shock filters. In: Michaelis B, Krell G, editors. Pattern recognition. Heidelberg: Springer; 2003. p. 1–8. [46] Coupe P, Hellier P, Kervrann C, Barillot C. Nonlocal meansbased speckle filtering for ultrasound images. IEEE Trans Image Process 2009;18(10):2221–9. [47] Deledalle CA, Denis L, Tupin F. Iterative weighted maximum likelihood denoising with probabilistic patch-based weights. IEEE Trans Image Process 2009;18(12):2661–72. [48] Ghosh S, Chaudhary KN. On fast bilateral filtering using Fourier kernels. IEEE Signal Process Lett 2016;23(5):570–3. [49] Sheet D, Pal S, Chakraborty A, Chatterjee A, Ray AK. Image quality assessment for performance evaluation of despeckle filters in optical coherence tomography of human skin. Proceedings of 2010 EMBS Conference on Biomedical Engineering and Sciences; 2010. [50] Wang Z, Bovik AC, Sheikh HR, Simoncelli EP. Image quality assessment: from error visibility to structural similarity. IEEE Trans Image Process 2004;3(4):1–14. [51] Wang Z, Bovik AC. A universal image quality index. IEEE Signal Process Lett 2002;9(3):81–4. [52] Norouzi A, Mohd Rahim MS, Altameem A, Saba T, Rad AE, Rehman A, et al. Medical image segmentation methods, algorithms, and applications. IETE Techn Rev 2014;31(3):199–213. [53] Horsch K, Giger ML, Venta LA, Vyborny CJ. Automatic segmentation of breast lesions on ultrasound. Med Phys 2001;28(8):1652–9. [54] Shan J, Cheng HD, Wang Y. A novel segmentation method for breast ultrasound images based on neutrosophic Imean clustering. Med Phys 2012;39(9):5669–82. [55] Liu B, Cheng HD, Huang J, Tian J, Tang X, Liu J. Probability density difference-based active contour for ultrasound image segmentation. Pattern Recogn 2010;43(6):2028–42. [56] Gao L, Liu X, Chen W. Phase and GFV-based level set segmentation of ultrasonic breast tumors. J Appl Math 2012;2012:1–22. [57] Xiang Y, Chung ACS, Ye J. An active contour model for image segmentation based on elastic interaction. J Comput Phys 2006;219:455–76.

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002

641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709

BBE 310 1–17 biocybernetics and biomedical engineering xxx (2018) xxx–xxx

710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742

Q2

[58] Mahesan KV, Bhargavi S, Jayadevappa D. Segmentation of MR images using active contours: methods, challenges and applications, 2017; 2 (4): 13–21. [59] Lotfollahi M, Gity M, Ye JY, Mahlooji Far A. Segmentation of breast ultrasound images based on active contours using neutrosophic theory. J Med Ultrason 2017. doi: 10.1007/ s10396-017-0811-8. [60] Lu R, Shen Y. Automatic ultrasound image segmentation by active contour model based on texture. Proceedings of the 1st International Conference on Innovative Computing Information and Control; 2006. [61] Li C, Huang R, Ding Z, Gatenby JC, Metaxas DN, Gore JC. A level set method for image segmentation in the presence of intensity inhomogeneities with application to MRI. IEEE Trans Image Process 2011;20(7):2007–16. [62] Sachdeva J, Kumar v, Gupta I, Khandelwal N, Ahuja CM. A novel content-based active contour model for brain tumor segmentation. Med Resonance Imaging 2012;30 (5):694–715. [63] Qian X, Wang J, Guo S, Li Q. An active contour model for medical image segmentation with application to brain CT image. Med Phys 2013;40(2). 021911-1-021911-10. [64] Yazdanpanah A, Hamarnesh G, Smith BR, Sarunic MV. Segmentation of intra-retinal layers from optical coherence tomography images using an active contour approach. IEEE Trans Med Imag 2011;30(2):484–96. [65] Shah S, Abaza A, Ross A, Ammar H. Automatic tooth segmentation using active contour without edges. 2006 Biometrics Symposium: Special Session on Research at the Biometric Consortium; 2006. [66] Middleton I, Damper RI. Segmentation of magnetic resonance images using a combination of neural networks and active contour models. Med Eng Phys 2004;26(1):71–86.

17

[67] Chalana V, Linker DT, Haynor DR, Kim Y. A multiple active contour model for cardiac boundary detection on echocardiographic sequences. IEEE Trans Med Imag 1996;15 (3):290–8. [68] Abdoli M, Dierckx RAJO, Zaidi H. Contourlet-based active contour model for PET image segmentation. Med Phys 2013;40(8). http://dx.doi.org/10.1118/1.4816296 [69] Derraz F, Beladgham M, Khelif M. Application of active contour models in medical image segmentation. Proceedings of the 2004 International Conference on Information Technology: Coding and Computing; 2004. [70] Atkins MS, Mackiewich BT. Fully automatic segmentation of the brain in MRI. IEEE Trans Med Imag 1998;17(1):98–107. [71] Klemencic A, Kovacic S, Pernus F. Automated segmentation of muscle fiber images using active contour models. Cytometery 1998;32(4):317–26. [72] Davayzikos CA, Prince JL. An active contour model for mapping the cortex. IEEE Trans Med Imag 1995;14(1):65–80. [73] Wang L, Li C, sun Q, Xia D, Kao CY. Active contours driven by local and global intensity fitting energy with application to brain MR image segmentation. Comp Med Imaging Graph 2009;33(7):520–31. [74] Al-Diri B, Hunter A, Steel D. An active contour model for segmenting and measuring retinal vessels. IEEE Trans Med Imaging 2000;28(9):1488–97. [75] Casselles V, Kimmel R, Sapiro G. Geodesic active contours. Int J Comp Vision 1997;22(1):61–79. [76] Chan TF, Vese LA. Active contours without edges. IEEE Trans Image Process 2001;10(2):266–77. [77] Image Processing and Analysis in JAVA, Image J 1.49 version 1.6.024 http://imagej.nih.gov/ij/download/win32/ij149-jre6-64.zip.

743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776

Please cite this article in press as: Nonogi H. The necessity of conversion from coronary care unit to the cardiovascular intensive care unit required for cardiologists. Biocybern Biomed Eng (2018), https://doi.org/10.1016/j.bbe.2018.10.002