Computers in Biology and Medicine 53 (2014) 220–229

Contents lists available at ScienceDirect

Computers in Biology and Medicine journal homepage: www.elsevier.com/locate/cbm

Manual and automated intima-media thickness and diameter measurements of the common carotid artery in patients with renal failure disease Christos P. Loizou a,b,n, Takis Kasparis b, Theodoros Lazarou c, Constandinos S. Pattichis d, Marios Pantziaris e a

Department of Computer Science, Intercollege, P.O. Box 51604, CY-3507 Limassol, Cyprus Cyprus University of Technology, Department of Electrical, Computer Engineering and Informatics, Limassol, Cyprus Nephrology Clinic, Limassol General Hospital, Limassol, Cyprus d Department of Computer Science, University of Cyprus, Nicosia, Cyprus e Cyprus Institute of Neurology and Genetics, Nicosia, Cyprus b c

art ic l e i nf o

a b s t r a c t

Article history: Received 3 March 2014 Accepted 1 August 2014

The objective of this study was to investigate differences in intima-media thickness (IMT) and diameter (D) measurements of the common carotid artery (CCA) in ultrasound imaging in normal subjects and renal failure disease (RFD) patients. Manual measurements by two experts and automated segmentation measurements (based on snakes and active contour models (ACM)) were carried out on 73 normal subjects, and 80 RFD patients. Statistical analysis was carried out using the Wilcoxon rank-sum test at po 0.05. Results demonstrated that the mean IMT and D measurements were significantly higher for the RFD group versus the normal group. Moreover, there was no significant difference between the manual and automated measurements. The ACM segmentation was slightly more accurate than segmentation based on snakes. Further work is needed to validate these findings on a larger group of subjects. & 2014 Elsevier Ltd. All rights reserved.

Keywords: Intima-media thickness Carotid diameter Renal failure disease Ultrasound imaging Common carotid artery Segmentation

1. Introduction The build-up on the artery walls known as atherosclerosis leads to cardiovascular disease (CVD) and results to heart attack and/or stroke. Atherosclerosis is widely-accepted as the main cause of morbidity and mortality to patients who undergo hemodialysis (HD) with end-stage renal failure disease (RFD) [1–3]. The distance between the lumen-intima and the media-adventitia is known as the intima-media thickness (IMT) of the common carotid artery (CCA). It can be determined by measuring the thickness of the innermost two layers of the arterial wall using non-invasive B-mode ultrasound imaging [2]. The IMT is well-accepted as a validated surrogate marker either for the study of atherosclerosis disease or for the study of endothelial dependent function [2]. It was shown that RFD patients undergoing HD, compared to the age- and gender-matched normal controls exhibit increased carotid IMT values [3], impaired endothelium dependent, but

n Corresponding author at: Intercollege, 92 Ayias Phylaxeos Street, P.O. Box 51604, CY-3507 Limassol, Cyprus. Tel.: þ 357 25 381180; fax: þ 357 25 386982. E-mail addresses: [email protected], [email protected] (C.P. Loizou), [email protected] (T. Kasparis), [email protected] (T. Lazarou), [email protected] (C.S. Pattichis), [email protected] (M. Pantziaris).

http://dx.doi.org/10.1016/j.compbiomed.2014.08.003 0010-4825/& 2014 Elsevier Ltd. All rights reserved.

unimpaired endothelium independent dilatation [4], and a higher risk of coronary artery disease [5]. Moreover, it was documented in [6,7], that the vascular walls of RFD patients are affected by traditional risk factors for increased atherosclerosis, such as dyslipidemia, diabetes mellitus, and hypertension. In all of the above studies, the intima-media complex (IMC) was segmented manually by experts. This is a time consuming process [8–11] and automated segmentation is highly desirable, especially when the image database is large [9–11]. Automated segmentation measurements may also reduce the intra- and inter-observer variability [8–11]. IMT can be measured through segmentation of the IMC, which corresponds to the intima- and media-layers (see also Fig. 1) of the arterial CCA wall. A number of techniques for the automated segmentation of the IMC from ultrasound images of the CCA have been proposed in the literature [8]. Recently, in [9,10] we presented a semi-automated IMC segmentation method based on normalization, despeckle filtering and snake-based segmentation which was also used in [9–13]. The system proposed in [10] is an extension of the system in [9] so that the intima- and media-layers of the CCA could also be segmented. A fully automated IMC segmentation algorithm for ultrasound images, based on active contours [14] and active contours without edges [15] (ACM) was presented in [11].

C.P. Loizou et al. / Computers in Biology and Medicine 53 (2014) 220–229

221

Fig. 1. An ultrasound image acquired from an RFD male patient at the age of 54. IMC and D manual segmentation measurements from the, (a) First expert (1) and (b) Second expert (2) (IMTmean1 ¼ 0.73 mm, IMTmax1 ¼ 0.85 mm, IMTmin1 ¼ 0.55 mm, IMTmedian1 ¼ 0.66 mm, Dmean1 ¼ 5.81 mm, Dmax1 ¼ 5.95 mm, Dmin1 ¼ 4.61 mm, Dmedian1 ¼5.73 mm), (IMTmean2 ¼ 0.70 mm, IMTmax2 ¼ 0.83 mm, IMTmin2 ¼ 0.59 mm, IMTmedian2 ¼0.67 mm, Dmean2 ¼5.97 mm, Dmax2 ¼ 6.12 mm, Dmin2 ¼4.68 mm, Dmedian2 ¼5.78 mm), (c) Automated IMC and D snakes segmentation measurements (IMTmean ¼ 0.71 mm, IMTmax ¼ 0.83 mm, IMTmin ¼ 0.51 mm, IMTmedian ¼ 0.67 mm, Dmean ¼ 5.71 mm, Dmax ¼ 5.75 mm, Dmin ¼ 4.69 mm, Dmedian ¼ 5.75 mm) and (d) Automated IMC and D active contour model segmentation measurements (IMTmean ¼ 0.70 mm, IMTmax ¼ 0.81 mm, IMTmin ¼ 0.49 mm, IMTmedian ¼0.64 mm, Dmean ¼5.66 mm, Dmax ¼5.69 mm, Dmin ¼ 4.67 mm, Dmedian ¼5.72 mm).

A number of studies which investigated the effect of increased IMT in the CCA and its relation to the progression of RFD in HD patients have been performed in the literature. In these studies the segmentations were performed manually by experts. More specifically, in [16], CCA and brachial IMT were evaluated in diabetic and non-diabetic RFD patients undergoing HD. In [5], a correlation between the IMT and age in RFD patients was found, while in [17], kidney dysfunction was associated with carotid atherosclerosis in patients with mild or moderate chronic RFD. Increased IMT values in chronic RFD patients were also reported in [18], while in [19] it was found that HD pediatric patients had reduced endothelial function with a development of carotid arteriopathy. In [20] it was shown that CCA IMT reflects the risk for ischemic heart disease in HD patients. The objective of this study was to investigate differences in IMT and D measurements of the CCA in ultrasound imaging in normal subjects and RFD patients. Manual measurements by two experts and automated segmentation measurements (based on snakes and ACM) were carried out on 73 normal subjects, and 80 RFD patients. Although the two systems were previously described in [9–13], in this work we expand our methodology to include the

measurements of the CCA diameter, D. Furthermore, we have evaluated the system using a new image database not previously available and we report the findings. Our study showed that the IMT and D measurements are larger in RFD patients than in normal individuals. We found significant statistical differences of the IMT and D measurements between the two groups (normal and RFD) but no significant statistical differences between the automated and the manual segmentation measurements. Both segmentation systems had similar performance while the ACM segmentation system produced measurements which were slightly closer to manual delineations made by the experts. The findings of this study may be helpful in understanding the development of RFD and its correlation with atherosclerosis in HD patients.

2. Materials and methods 2.1. Recording of ultrasound images A total of 80 B-mode longitudinal ultrasound images of the CCA from RFD patients, displaying the vascular wall as a regular pattern

222

C.P. Loizou et al. / Computers in Biology and Medicine 53 (2014) 220–229

(see Fig. 1a) that correlates with anatomical layers were recorded (task carried out by co-authors T. Lazarou and M. Pantziaris). The RFD patients, 36 women and 44 men, had a mean 7 SD age of 63.0776.82 years (5 patients had CVD symptoms). Additionally, 73 ultrasound B-mode images from healthy individuals, (24 women and 49 men), were acquired, at a mean 7SD age of 61.78 711.45 years. The images were acquired using the ATL HDI-5000 ultrasound scanner (Advanced Technology Laboratories, Seattle, USA) [21] with a resolution of 576  768 pixels with 256 gray levels. The spatial gray resolution used for the image acquisitions was 17 pixels per mm (i.e. the resolution is 60 mm). The ATL HDI-5000 ultrasound scanner was further equipped with a 256-element fine pitch high-resolution 50-mm linear array, a multi-element ultrasound scan head with an extended operating frequency range of 5–12 MHz and it offers real spatial compound imaging. For the recordings, a linear probe (L74) at a recording frequency of 7 MHz was used. Assuming a sound velocity propagation of 1550 m/s and 1 cycle per pulse, we thus have an effective spatial pulse width of 0.22 mm with an axial system resolution of 0.11 mm [21]. We use bicubic spline interpolation to resize all images to a standard pixel density of 16.66 pixels/mm (with a resulting pixel width of 0.06 mm) [22]. A written informed consent from each subject was obtained according to the regulations of the local ethics committee. 2.2. Ultrasound image normalization In order to reduce variability introduced by different gain settings, different operators and different equipment, and improve ultrasound tissue comparability, the ultrasound images were brightness adjusted based on the method described in [22] and [9–13] (see also Fig. 1). Algebraic (linear) scaling of the images were manually performed by linearly adjusting the image so that the range of the median gray level values of the blood was 0–5, and the adventitia (artery wall) was 180–190 [22]. The scale of the gray level of the images ranged from 0 to 255. Thus the brightness of all pixels in the image was readjusted according to the linear scale defined by selecting the two reference regions. Fig. 1a and b presents manual IMC and D segmentations (see also Section 2.3) performed by both experts respectively. Fig. 1c and d illustrates the automated IMC and D segmentations based on snakes [23] and ACM [11]. The segmentations were performed after image normalization and speckle reduction filtering (see also Section 2.4). It is noted that a key point to maintaining a high reproducibility was to ensure that the ultrasound beam was at right angles to the adventitia, which it was visible adjacent to the plaque. 2.3. Manual IMC and diameter segmentation A neurovascular expert (co-author M. Pantziaris) and a vascular expert (co-author T. Lazarou) with more than 30 years of clinical experience, manually delineated the IMC and the carotid diameter, D [9–10], on 80 RFD longitudinal ultrasound images of the CCA after image normalization (see Section 2.2) and speckle reduction filtering (see Section 2.4). The first expert also manually delineated the IMC and D on 73 ultrasound images of the CCA acquired from normal subjects. The IMT was measured by selecting 20–30 consecutive points for the intima and the adventitia layers at the far wall of the CCA, while the diameter was measured by selecting 20–30 consecutive points at the near and far walls of the CCA. The manual delineations were made using a system implemented in MATLABs (Math Works, Natick, MA) by our group. The measurements were performed between 1 and 2 cm proximal to the bifurcation of the CCA on the far wall [22], over a distance of 1.5 cm starting at a point 0.5 cm and ending at a point 2.0 cm proximal to the carotid bifurcation. The bifurcation of the CCA was

used as a guide, and all measurements were made from that region. The IMT and the diameter, D, were then calculated as the average of all the measurements. The measuring points and delineations were saved to be compared with the two segmentation methods. All sets of manual segmentation measurements were performed by the experts in a blind manner, both with respect to identifying the subject and to delineating the image by using the mouse. 2.4. Speckle reduction filtering (DsFlsmv) The filter DsFlsmv (despeckle filter linear scaling mean variance-DsFlsmv), introduced in [23], and evaluated in [9–13], was applied to image for speckle reduction, prior to the IMC and D segmentations. Filters of this type utilize first order statistics such as the variance and the mean of a pixel neighborhood and may be described with a multiplicative noise model [12,13,23]. Hence, algorithms in this class may generally be described by the following equation: f i;j ¼ g þ ki;j ðg i;j  gÞ

ð1Þ

where f i;j is the estimated noise-free pixel value, g i;j is the noisy pixel value in the moving window, g is the local mean value of an N1  N 2 region surrounding and including pixel g i;j , ki;j is a weighting factor, with k A ½0; 1, and i, j are the pixel coordinates. The factor ki;j is a function of the local statistics in a moving window and is defined [12,13,23] as ki;j ¼

ð1  g 2 σ 2 Þ : ðσ 2 þ σ 2n Þ

ð2Þ

The values, σ 2 and σ 2n , represent the variance in the moving window and the variance of the noise in the whole video frame respectively. The noise variance may be calculated for the logarithmically compressed image by computing the average noise variance over a number of windows with dimensions considerable larger than the filtering window [12,13,23]. The moving window size for the despeckle filter DsFlsmv was 5  5 and the number of iterations applied to each image was two. 2.5. IMC and diameter contour initialization Before running the automated segmentation algorithms, IMC and D initialization procedures [9,10,23] were carried out for positioning the initial snake contours as close as possible to the area of interest. The snake initialization procedure is described as follows: 1) Load the B-mode image (see Fig. 1a), perform image normalization (see Section 2.2) and despeckle filtering (see Section 2.4). A normalized despeckled image of the CCA is shown in Fig. 1a and b. 2) Binarize the image in order to extract edges more easily. 3) Erode the binary image (from point 2) above by applying an erode morphological operation that eliminates smaller not connected binary image areas. 4) Perform edge detection on the binary eroded image and extract the contour matrix of the image. Extract the contour matrix of the above area by locating points and their coordinates (contours) for the near and far walls of the CCA and construct two interpolating B-splines (adventitia for the near wall and adventitia for the far wall of the CCA). Sample the interpolating B-splines in 30 equal segments, in order to define 30 snake points on each contour. 5) Map the detected near wall and far wall adventitia contour points from point 4) above, on the image to form the initial snake contours for the near and far wall adventitia borders.

C.P. Loizou et al. / Computers in Biology and Medicine 53 (2014) 220–229

6) Displace the initial contour estimated in 4) for the far wall adventitia, upwards for up to 17 pixels (1.02 mm) to form the initial snake contour for the intima layer. This displacement is based on the observation that the IMT lies between 0.6 and 1.4 mm (0.6 mm oIMT o1.4 mm), with a mean IMT of 1.0 mm [25]. Taking into account that the spatial resolution (distance between 2 pixels) is 0.06 mm, then the IMT is lying within the range of 10 oIMTo24 pixels, with a mean of 17 pixels. Therefore the displacement of the contour, in order to estimate the intima should be in average 17 pixels. 7) Deform the initial contours by the snake and the ACM (see Section 2.6) to accurately locate the adventitia and intima borders for the far wall and the intima border for the near wall [9,10], as well as measure the CCA diameter [23], and save them. 8) Save and display the detected contours (i.e. far wall adventitia and intima and near wall intima) on the B-mode image (see Fig. 1c and d).

2.6. IMC and diameter segmentation The IMC and the D, of the CCA, were automatically segmented after image normalization and despeckle filtering (see Sections 2.2 and 2.4), using two different automated systems developed in MATLABs by our group and evaluated on 100 ultrasound images as reported in [9–11] respectively. For the first automated system the segmentations were based on the Williams and Shah [26] snake segmentation method, which was used to deform the snake and segment the IMC and the near wall borders in each image. The snake contour, vðsÞ, adapts itself by a dynamic process that minimizes an energy function (Esnake ðv; sÞ) defined as [26] Esnake ðvðsÞÞ ¼ Eint ðνðsÞÞ þ Eimage ðvðsÞÞ þ Eexternal ðvðsÞÞ Z ¼ ðαs Econt ðvðsÞÞ þ βs Ecurv ðvðsÞÞ þ γ s Eimage ðvðsÞÞ s

þ Eexternal ðvðsÞÞÞds:

ð3Þ

where Eint ðvðsÞÞ, Eimage ðvðsÞÞ, Eexternal ðvðsÞÞ, Econt ðvðsÞÞ, and Ecurv ðvðsÞÞ are the internal, image, external, continuity, and curvature energies of the snake, and αs , βs and γ s the strength, tension and stiffness parameters respectively. For the Williams & Shah method, the strength, tension and stiffness parameters were equal to αs ¼ 0:6, βs ¼ 0:4, and γ s ¼ 2 respectively [9–12]. The second automated system utilizes active contours [14], and active contours without edges [15]. Active contours without edges [15] are used to identify an initial intima-media boundary in each ultrasound image of the CCA and segment the image into lumen and carotid wall. The image is separated into regions based on pixel intensities and uses the following energy functional: Fðc1 ; c2 ; CÞ ¼ μ:LengthðCÞ þ ν:AreaðinsideðCÞÞ þ λ1

þ λ2

Z

insideðCÞ

ju0 ðx; yÞ c1 j2 dx dy

Z otsideðCÞ

ju0 ðx; yÞ  c2 j2 dx dy:

ð4Þ

where m Z0, ν Z 0, λ1, λ2 40, are fixed parameters. The segmentation problem becomes the minimization of (4) where all the details about the model and the selection of the parameters are given in [11]. The extracted final snake and ACM contours (see Fig. 1c and d), correspond to the adventitia and intima borders of the IMC at the far wall and the intima border at the near wall of the CCA as well as the diameter of the CCA. The distance is computed between the two boundaries (at the far wall), at all points along the arterial wall segment of interest moving perpendicularly between pixel pairs, and then averaged to obtain the mean IMT (IMTmean). Also

223

the maximum (IMTmax, Dmax), minimum (IMTmin, Dmin), and median (IMTmedian, Dmedian) IMT and D values, were calculated. The 80 ultrasound images acquired from the RFD patients as well as the 73 images acquired from normal subjects were automatically segmented using both segmentation techniques proposed in this study. 2.7. Statistical analysis The Wilcoxon rank-sum test [27,28] was used in order to identify if a significant difference (S) or not (NS) exists between the extracted IMC and D segmentation measurements. This was done for each set of measurements (normal subjects versus the RFD subjects) for independent samples of same sizes, for the manual and automated segmentation measurements, with a confidence level of 95%. For significant differences, we require po 0.05. For independent samples of different sizes, the Mann– Whitney rank-sum test [28] was used for detecting significant differences between male and female subjects. Furthermore, box plots for the IMT between the different groups, were plotted. Bland–Altman plots [27], with 95% confidence intervals, were also used to further evaluate the IMT and D segmentation measurements between the normal subjects and the RFD patients. Also, Spearman's correlation coefficient, ρ, between the normal and the RFD, IMT and D segmentation measurements, which reflects the extent of a linear relationship between two data sets, was investigated. We furthermore, use regression analysis to investigate the relationship between the IMT and age for the normal and RFD subjects.

3. Results Fig. 1 presents an ultrasound image of the CCA from a 54 year old male RFD patient. The IMC at the far wall and the D of the CCA were segmented manually (see Fig. 1a and b) and automatically (see Fig. 1c and d) after normalization (see also Section 2.2) and despeckle filtering (see also Section 2.4). More specifically, Fig. 1a and b presents the manual IMC and D segmentations from the first expert (1) and the second expert (2) respectively. In Fig. 1c we illustrate the automated IMC and D snakes segmentations, while Fig. 1d illustrates the automated ACM segmentations. We may observe that the IMT and D segmentation measurements are very close for both manual and automated segmentation methods. Furthermore, the ACM generates generally, lower values for the IMT and D when compared to the snake based segmentation method. In the left part of Table 1, we present the manual and the automated mean, maximum, minimum and median, IMT and D segmentation measurements in millimeters. The measurements were automatically performed on all ultrasound images acquired from the normal and RFD subjects, using both segmentation systems presented in this study. We may observe that both IMT and D manual and automated segmentation measurements, for the RFD patients, are higher when compared to the IMT and D segmentation measurements performed on normal subjects. Manual and automated IMT and D segmentation measurements are very close. Both segmentation systems performed satisfactorily with similar variability for both the IMT and the D measurements. In the right part of Table 1 we present the Mann–Whitey rank sum test performed between the male and the female subjects for the mean IMT and D segmentation measurements. No statistically significant differences were found for the normal as well as for the RFD patients for both manual and automated IMT and D segmentation measurements between the male and female subjects.

224

C.P. Loizou et al. / Computers in Biology and Medicine 53 (2014) 220–229

Table 2 tabulates the results of the statistical analysis for the normal and the RFD subjects for both the manual and automated IMT measurements. After performing the non-parametric Wilcoxon rank sum test presented in Table 2, we found no statistical significant differences (NS) between the two manual (p¼ 0.06) measurements for the RFD patients. We also found no statistical significant differences for the first manual as well as for the second manual measurements and the automated snakes and automated ACM IMT segmentation measurements (p ¼0.11 and p¼ 0.21 respectively). We have furthermore estimated strong correlation between the two manual IMT measurements (ρ ¼0.88) for the RFD patients. Also strong correlations were found between the automated snakes and the two manual IMT segmentation measurements (ρ ¼0.72 and ρ ¼0.98 for the first and the second manual measurements respectively) for the RFD patients. The correlation coefficient, ρ, between the ACM segmentation and the first manual measurements was ρ ¼ 0.89, while ρ between the ACM segmentation and the second manual measurements was ρ ¼0.96. The IMT

Table 1 Manual and automated mean, maximum, minimum 7 (standard deviation) and median (inter-quartile range, IQR) segmentation measurements for the IMC and diameter (D), for the 73 normal subjects and 80 RFD patients investigated in this study. Values are in mm. The right part of table presents the statistical Mann– Whitney rank sum test at po 0.05 between the male and the female subjects for the mean IMT. CCA IMT and Diameter measurements Mean [mm]

Maximum [mm]

Minimum [mm]

Median [mm]

Male versus Female

Normal subjects IMTman_1 0.73 IMTauto_s 0.74 IMTauto_acm 0.71 Dman_1 4.64 Dauto_s 4.76 Dauto_acm 4.71

(0.20) (0.21) (0.21) (0.99) (0.85) (0.84)

0.85 0.87 0.84 4.82 4.99 4.94

(0.27) (0.26) (0.25) (0.91) (0.92) (0.91)

0.54 0.56 0.53 4.49 4.53 4.48

(0.25) (0.23) (0.23) (1.19) (1.05) (1.05)

0.72 0.74 0.71 4.58 4.69 4.66

(0.26) (0.21) (0.21) (0.86) (0.76) (0.75)

NS NS NS NS NS NS

(0.45) (0.57) (0.51) (0.69) (0.57) (0.33)

RFD patients 0.92 IMTman_1 IMTman_2 0.96 IMTauto_s 0.96 IMTauto_acm 0.92 Dman_1 5.72 Dman_2 6.08 Dauto_s 6.19 Dauto_acm 6.13

(0.15) (0.14) (0.13) (0.13) (1.11) (0.63) (0.73) (0.72)

1.16 1.14 1.15 1.09 5.91 6.29 6.32 6.27

(0.22) (0.24) (0.24) (0.22) (1.14) (0.61) (0.69) (0.69)

0.81 0.76 0.76 0.73 5.51 5.86 5.83 5.78

(0.19) (0.23) (0.23) (0.22) (1.36) (0.67) (0.70) (0.71)

0.91 0.95 0.94 0.91 5.59 5.93 5.99 5.94

(0.15) (0.12) (0.12) (0.12) (1.34) (0.61) (0.68) (0.68)

NS NS NS NS NS NS NS NS

(0.06) (0.29) (0.26) (0.19) (0.22) (0.13) (0.66) (0.49)

IMTman_1, IMTman_2, Dman_1, Dman_2: Manual IMT and D segmentation measurements from the first and second expert. IMTauto_s, IMTauto_acm, Dauto_s, Dauto_acm: Automated IMT and D segmentation measurements for the snakes and the ACM segmentation methods respectively. The indices 1 and 2 indicate manual segmentation measurements from the first and the second expert respectively. NS: Nonsignificantly different.

measurements computed by the two different segmentation methods were found not significantly different (p ¼0.14), while a strong correlation coefficient was estimated between the IMT mean values obtained by the two different segmentation methods (ρ ¼ 0.88). The present findings indicate a strong relationship between the manual and the automated IMT segmentation measurements as well as between the two different automated segmentation methods. It is believed that the proposed segmentation systems may perform as reliably as the manual and that both, the snakes and the ACM segmentation systems may be reliably used for the segmentation of the CCA. When comparing the IMT normal measurements versus the RFD measurements (see the right part of Table 2), we found significant statistical differences as well as low correlation coefficient, (p¼ 0.0001, ρ ¼ 0.002 and p ¼0.001, ρ ¼ 0.01) for the first expert and the second expert respectively. The IMT automated snake and ACM segmentation measurements performed between the RFD versus normal subjects were also found to be significantly different (p ¼0.001, p ¼ 0.01), with low correlation coefficient (ρ ¼ 0.001, ρ ¼0.09) respectively. Similar findings were also found in this study for the diameter, D of the CCA. Fig. 2 presents box plots for the mean manual and automated segmentation measurements of the CCA IMT (see Fig. 2a) and the D (see Fig. 2b), performed in this study. From left to right, we observe in Fig. 2a and b, the mean manual IMT and D for the two experts (IMT_m1, IMT_m2, D_m1, D_m2), the mean automated snakes and ACM IMT and D (IMT_snake, IMT_acm, D_snake, D_acm) on RFD patients, and the mean manual from first expert (IMTnor_m1, Dnor_m1) and automated (IMTnor_acm, Dnor_acm) IMT and D on normal subjects respectively. We may observe that for both the IMT and the D segmentation measurements of the CCA, the normal values are lower than those of the RFD patients for both the manual and the automated segmentation measurements. Manual and automated IMT and D segmentation measurements are close. The two different automated segmentation methods performed similarly in terms of the IMT and D measurements. Fig. 3a illustrates Bland–Altman plots for the mean IMT measurements, of the RFD and normal subjects investigated in this study, for the manual (measured by the first expert, see Fig. 3a) and the automated (measured with the ACM method, see Fig. 3b) segmentation measurements respectively. In Fig. 3a we may observe a difference between the normal and RFD IMT manual measurements of 0.20 þ0.66 mm and 0.20–0.26 mm. In Fig. 3b we have for the automated measurements a difference between the normal and RFD of 0.22 þ0.71 mm and 0.22–0.27 mm respectively. Fig. 3c and d presents regression plots for the mean normal IMT (segmented by the ACM) versus age (see Fig. 3c), and the mean RFD IMT (segmented with snakes) versus age (see Fig. 3d). The ACM automated mean IMT for the normal subjects as a function of age (see Fig. 3c) has an estimated intercept of

Table 2 Wilcoxon rank-sum test performed at p o 0.05 and the Spearman's correlation coefficient, ρ, for the IMT segmentation measurements investigated in this study. p-values are given in parentheses. S and NS indicate significantly and non-significantly difference at p o 0.05 and p 4 ¼ 0.05 respectively. IMT Segmentation measurements performed on RFD patients

Manual_1 Manual_2 Automated-snakes

RFD patients

Normal subjects

Manual_2

Automated-snakes

Automated-ACM

Manual_1

NS (0.06) ρ ¼ 0.88 (p ¼ 0:001)

NS (0.11) ρ ¼0.72 (p ¼ 0:99) NS (0.21) ρ ¼0.98 (p ¼ 0:001)

NS (0.06) ρ ¼0.89 (p ¼ 0:001) NS (0.16) ρ ¼0.96 (p ¼ 0:001) NS (0.14) ρ ¼0.88 (p ¼ 0:01)

S (0.0001) ρ ¼0.002 (p ¼ 0:99) S (0.001) ρ ¼0.01 (p ¼ 0:93) S (0.001) ρ ¼0.001 (p ¼ 0:98) S (0.01) ρ ¼0.09 (p ¼ 0:81)

Automated-ACM

RFD: Renal failure disease, Manual_1, Manual_2: Manual segmentation measurements from expert 1 and expert 2, ACM: Active contour model.

C.P. Loizou et al. / Computers in Biology and Medicine 53 (2014) 220–229

225

Fig. 2. Box plots for the mean segmentation measurements of the CCA for: (a) IMT, and (b) Diameter (D) of the CCA. From left to right, RFD patients' mean manual IMT and D segmentation measurements for the two experts (IMT_m1, IMT_m2, D_m1, D_m2), mean automated IMT and D for the two different segmentation methods (IMT_snake, IMT_acm, D_snake, D_acm,) and normal subjects' mean manual and ACM automated IMT and D on normal subjects (IMTnor_m1, IMTnor_acm, Dnor_m1, Dnor_acm) respectively. Inter-quartile range (IQR) values are shown above the box plots. Straight lines connect the nearest observations with 1.5 of the IQR of the lower and upper quartiles. Unfilled rectangles indicate possible outliers with values beyond the ends of the 1.5xIQR.

0.6121 (standard error of 0.1349) and a slope of 0.001433 (standard error of 0.0021) with a correlation coefficient, ρ ¼0.079 (p ¼0.5067). We estimated an F-ratio of 0.2345. Fig. 3d shows that the snakes automated mean IMT for the RFD subjects has an estimated intercept of 0.7092 (standard error of 0.138) and a slope of 0.00386 (standard error 0.0022) with a correlation coefficient, ρ ¼ 0.20 (p ¼0.29). We estimated an F-ratio of 3.21. The results provide quantitative evidence of no linear relationship between the IMT and age for both normal and RFD patients in contrast to a linear relationship reported in other studies [18,20,29,30]. When comparing Fig. 3c with Fig. 3d we may observe that a stronger correlation coefficient exists between the IMT and age for the RFD patients as well as a stronger increase of the IMT with age when compared with that of the normal population. The findings of

Fig. 3c and d are also applicable in the case of manual IMT measurements' relationship with age.

4. Discussion The results of this study showed that the IMT and D segmentation measurements of the CCA in RFD patients are higher than the values found in the normal subjects (see Table 1, Figs. 2 and 3b and c). The study also showed that there are statistical significant differences for both the IMT and D segmentation measurements between the normal subjects and the RFD patients (see also Table 2). For both segmentation methods investigated in this

226

C.P. Loizou et al. / Computers in Biology and Medicine 53 (2014) 220–229

Fig. 3. Regression analysis (Bland–Altman plots). The middle line represents the mean difference, and the upper and lower outside lines represent the limits of agreement between the two measurements, which are the mean of the data 7 2sd for the estimated difference between the two measurements. (a) Mean manual IMT (delineated by the first expert) between the normal and the RFD subjects, (b) Mean ACM automated IMT for the RFD patients and normal subjects. (c) Increase of mean automated IMT with age (estimated with the ACM method) for normal subjects, (d) Increase of mean automated IMT with age (estimated with the snakes method) for RFD patients with age.

study, we found no significant differences between the two manual and the two automated segmentation measurements for the IMT and the D of the CCA. Also, no statistical significant differences were found between the two segmentation methods for the measurements of the IMT and D (see also Table 2). The ACM segmentation system performed slightly better than the snakes based system. The findings of this study indicate that the two different segmentation systems, evaluated in this study, may be used to complement the manual IMC and D segmentation analysis in the assessment of RFD patients. The mean IMT and D measurements for the normal subjects, exhibit both lower values when compared to the corresponding values of the RFD patients (see Table 1). This is also the case for all the other measurements of the IMT (maximum, minimum, and median). Our findings show that kidney dysfunction may probably

thus be associated with increased carotid IMT and D. This is in accordance with other studies reported earlier in the literature [20,32–36], and reflects a progression of athero/arteriosclerotic changes following a natural ageing process. More specifically, in [34] increased IMT and carotid wall-to-lumen ratio, were found in 60 RFD patients, whereas in [35] it was shown that CCA IMT in 75 hemodialysis patients was significantly higher compared with 75 kidney transplant recipients and that it increased with a prolongation of dialysis duration. The authors in [35], reported a CCA IMT mean thickness of 1.270.35 mm for the hemodialysis patients and a mean IMT thickness of 0.73 70.18 mm for the kidney transplanted patients (po 0.001). Atherosclerosis is the most common cause of neurovascular disease, resulting in hypertension and ischemic nephropathy [2,6]. A regular follow up was suggested in [6], for selected patients with

C.P. Loizou et al. / Computers in Biology and Medicine 53 (2014) 220–229

atherosclerotic renal artery stenosis before they develop severe hypertension or renal insufficiency. Randomized clinical trials were also proposed to define the role of early renal revascularization in the management of these patients. An association between CCA IMT values and cardiovascular mortality in non-diabetic RFD patients was documented in [36]. It was shown that IMT may represent an independent predictor of cardiovascular mortality in these patients and IMT measurements may be used for cardiovascular mortality risk stratification also in the dialysis population. In another study [38], CCA IMT and carotid diameter were measured in 70 RFD patients and 50 normal subjects. Compared with normal subjects, RFD patients had increased CCA diameter (6.25 7 0.87 versus 5.55 70.65 mm; p o0.001), and larger CCA IMT (0.777 70.115 mm versus 0.67870.105 mm po 0.001). A small diameter reflects to a healthy vasculature that is able to maintain an optimal balance of shear and tensile stress [39]. An enlarged diameter is less able to effectively control the levels of shear stress. This can make the artery vulnerable to injury and atherosclerotic development [40]. Similar findings were also reported in the present study as well as in [41], where significantly higher IMT values were found compared to healthy controls (0.9 70.85 mm versus 0.6 70.55 mm). The authors in [38], concluded that IMT may be used as a predictor of mortality in chronic kidney disease. Different measures of cardiovascular risk stratification were evaluated in [42] such as coronary artery calcification, IMT, carotid plaque, and the Framingham risk score in 220 patients with chronic RFD. The authors identified a high burden of atherosclerosis among individuals with RFD as well as statistical differences between coronary artery calcification and carotid plaque or IMT as predictors of self-reported prevalent CVD. In a large study [29], 1351 male individuals were investigated in order to estimate whether chronic RFD was associated with carotid IMT thickening. It was shown that chronic RFD may be associated with early carotid atherosclerosis in low-risk individuals, such as those undergoing general health screening, who have hypertension and/or impaired glucose metabolism. It was also shown in [29], as well as in [9,18,20,30,32], that there was a linear correlation between the progression of the IMT and the age of the subjects. This is not the case in the present study where a low correlation coefficient for the IMT versus age for both the normal and the RFD patients was found. A linear association between IMT and age could not be established and this may be probably attributed to the small number of cases investigated as well as to the high age of normal subjects used. In [16] the authors evaluated the IMT in the CCA and in brachial arteries, flow-mediated dilatation (FMD), and nitroglycerinmediated dilatation (NMD) in diabetic and non-diabetic hemodialysis patients. A positive correlation was found, with respect to age, with carotid IMT and brachial IMT. Also a negative correlation was found with FMD and NMD in the cases of hypertension and diabetes. It was thus demonstrated that age was the most important factor that significantly affected all studied markers of atherosclerosis in hemodialysis patients. The authors in [5] found a correlation between IMT and age in RFD patients and that IMT values were correlated with total cholesterol. It was furthermore reported in [17], that kidney dysfunction was associated with increased carotid IMT independent of the classical atherosclerotic risk factors but they found no association between kidney dysfunction and calcified plaque. Carotid IMT may be a predictive marker of CVD that reflects early kidney dysfunction, especially in high-risk patients. Finally, in [30], the relation between carotid IMT and hemodialysis patients in chronic RFD was investigated, where 78 RFD patients were examined. A significant positive correlation was found between IMT and hemodialysis (p ¼0.045) independent of traditional risk factors. It was also shown that hemodialysis

227

patients had significantly higher mean IMT (1.136 70.21 mm) compared to that of normal (0.959 70.23 mm) subjects.

4.1. Limitations There are also some limitations for the present study which arise mostly from the relatively small number of subjects included in this study and that the duration of diabetes and RFD were not included. The material used in this study was acquired from RFD patients undergoing hemodialysis. Additionally, the study should be further applied on a larger sample, a task which is currently undertaken by our group. It seems convincible that IMT may reflect the risk in RFD subjects, but additional variables, such as age, sex, weight, blood pressure and others should be taken into account for better evaluating the RFD in hemodialysis patients. Furthermore, the presence of acoustic shadowing together with strong speckle noise hinders the visual and automatic analysis in ultrasound images. Such images, with bad visual perception, were neither included in this study nor were they delineated by the experts [8–13,31]. We have also excluded from our segmentation experiments images with extensive echolucency and calcification. Furthermore, the estimation and positioning of the initial snake contour may sometimes result to segmentation errors where the ACM showed to perform better. In cases where segmentation is performed by snakes, the initial contour should be placed as close as possible to the area of interest otherwise it may be trapped into local minima or false edges and converge to a wrong location. The ACM segmentation method [11] is fully automated compared to the snake's segmentation method in [9], it requires no user interaction or manual correction and the initial positioning of the snake is more accurate. The snake usually produces smoother boundaries; however the segmentation values obtained by the ACM method appear to be closer to manual (see also Table 1) although the variation for the two different segmentation methods is about the same. It should be noted that the parameters for each processing step (image normalization, despeckle filtering, the size of the moving pixel window, the number of iterations, IMC contour initialization, snakes segmentation) were selected for maximum performance. In the present study in less than 4% of the cases the positioning of the initial snake contour was not calculated correctly. Furthermore, only vessels without atherosclerotic plaques were segmented in this study. The average computation of the snake contours and the extraction of the IMC, and the D, of the CCA was about 16 s/image, using a Pentium III desktop computer with 3.2 GHz RAM memory, where the segmentation by the ACM required about 21 s. It is noted that the expert clinician required over 25–30 s/frame for the manual delineation. An average processing time of 20 s/image in case of the segmentation of the CCA images is reported in [32] (p. 1267, end of Section 2). In [33] the average processing time per image was 30 s whereas in [8] an average time of about 40 s was reported for a number of different automated segmentation methods. The proposed methods can also be incorporated into a system as a commercial version, while the proposed algorithms may be designed to perform parallel with the clinician's evaluation, so that the clinician might carry on the patient examination, while the IMC and D segmentation measurements are performed. Finally, it is noted that the IMT should be measured preferably on the far wall of the CCA as this will avoid inter-observer variability. Values from the near wall depend in part on gain settings and are less reliable [22,25,31]. In cases where IMT measurements are taken on the near wall, their values should be recorded separately from IMT of the far wall.

228

C.P. Loizou et al. / Computers in Biology and Medicine 53 (2014) 220–229

4.2. Future work Future work will investigate the possible incorporation of the proposed segmentation system into a computer aided diagnostic system that supports the texture image analysis of the segmented IMC areas, as documented in [32] providing an automated system for the early diagnosis and the assessment of the risk of stroke in patients undergoing hemodialysis. Future studies including a higher number of patients are necessary to determine the impact of different risk factors, especially non-traditional ones, on CVD and RFD disease in these patients. In the meantime, aggressive management to reduce these risk factors is a strategy for preventing CVD in RFD patients. Furthermore, it would be of interest to investigate the motion characteristics of the arterial wall in this group of patients by performing segmentation of the atherosclerotic carotid plaque in ultrasound videos as it was done in [43], where different motion and wall stress parameters from the atherosclerotic plaque and carotid wall can be extracted by utilizing the motion mode image of the CCA [44]. The use of despeckle filtering in ultrasound image [45] and video [46] analysis, could also be investigated for increasing the visual performance, and the manual as well as the automated video segmentation accuracy of the IMC and diameter of the CCA in this group of subjects. Finally, a new texture image retrieval system that uses texture features extracted from the IMC, to retrieve images that could be associated with the same level of the risk for developing RFD will be incorporated.

5. Concluding remarks This research work presents the application of two different segmentation systems for segmenting the IMC and D in ultrasound images of the CCA. Both systems can be used to complement manual measurements. The ACM based segmentation system performed slightly better than the snakes based segmentation system. It was also shown in this study, that the degree of atherosclerosis is greater in RFD patients when compared to that of normal subjects. We anticipate that the above comparison may have some clinical value in retrospectively screening of the dysfunction of the CCA through time. It would be furthermore interesting in the future to study the atherosclerotic carotid plaque changes in the aforementioned group of subjects and estimate whether and how kidney dysfunction is associated with the buildup of carotid plaque. It may also be possible to identify and differentiate those individuals into high and low risk groups according to their CVD risk before the development of the atherosclerotic carotid plaque. The use of texture features, as also proposed in [32], could also be used in order to provide new feature sets, which can be used in combination with IMT and D measurements for differentiating between normal and RFD IMC structure.

Conflict of interest statement None declared. References [1] S. Mendis, P. Puska, B. Norrving, Global Atlas on Cardiovascular Disease Prevention and Control, WHO, ISBN 978-92-4-156437-3, 2012. [2] R.N. Foley, P.S. Parfrey, M.J. Sarnak, Clinical epidemiology of cardiovascular disease in chronic renal disease, Am. J. Kidney Dis. 32 (1998) 112–119. [3] R. Hojs, Carotid intima-media thickness and plaques in hemodialysis patients, Artif. Organs 24 (2000) 691–695.

[4] C. van Guldener, J. Lambert, M.J. Janssen, A.J. Donker, C.D. Stehouwer, Endothelium-dependent vasodilatation and distensibility of large arteries in chronic hemodialysis patients, Nephrol. Dial. Transplant 12 (2) (1997) 14–18. [5] S. Bevc, R. Hojs, R. Eckart, T. Hojs-Fabjan, Atherosclerosis in hemodialysis patients: traditional and nontraditional risk factors, Acta Dermatoven APA 15 (4) (2006) 151–157. [6] E.R. Zielner, Atherosclerotic renovascular disease: natural history and diagnosis, Perspect. Vasc. Surg. Endovasc. Ther. 16 (4) (2004) 299–310. [7] N. Maeda, Y. Sawayama, M. Tatsukawa, et al., Carotid artery lesions and atherosclerotic risk factors in Japanese hemodialysis patients, Atherosclerosis 169 (2003) 183–192. [8] F. Molinari, G. Zeng, J.S. Suri, A state of the art review on intima-media thickness measurement and wall segmentation techniques for carotid ultrasound, Comp. Methods Progr. Biomed. 100 (2010) 201–221. [9] C.P. Loizou, C.S. Pattichis, M. Pantziaris, T. Tyllis, A.N. Nicolaides, Snakes based segmentation of the common carotid artery intima media, Med. Biol. Eng. Comput. 45 (2007) 35–49. [10] C.P. Loizou, C.S. Pattichis, A.N. Nicolaides, M. Pantziaris, Manual and automated media and intima thickness measurements of the common carotid artery, IEEE Trans. Ultrason. Ferroelectr. Freq. Control 56 (5) (2009) 983–994. [11] S. Petroudi, C.P. Loizou, M. Pantziaris, C.S. Pattichis, Segmentation of the common carotid intima-media complex in ultrasound images using active contours, IEEE Trans. Biomed. Eng. 59 (11) (2012) 3060–3069. [12] C.P. Loizou, C.S. Pattichis., Despeckle filtering algorithms and software for ultrasound imaging, Synthesis lectures on algorithms and software for engineering, Morgan & Claypool Publishers, USA, 2008. [13] C.P. Loizou, C.S. Pattichis, C.I. Christodoulou, R.S.H. Istepanian, M. Pantziaris, A. N. Nicolaides, Comparative evaluation of despeckle filtering in ultrasound imaging of the carotid artery, IEEE Trans. Ultrason. Ferroelectr. Freq. Control 52 (10) (2005) 1653–1669. [14] M. Kass, A. Witkin, D. Terzopoulos, Snakes: active contour models, Int. J. Comput. Vis. 1 (1988) 321–331. [15] T. Chan, L. Vese, Active contours without edges, IEEE Trans. Image Process. 10 (2) (2001) 266–277. [16] R. Rus, J. Butotovic-Ponikvar, Intima media thickness and endothelial function in chronic hemodialysis patients, Ther. Apher. Dial. 13 (4) (2009) 249–299. [17] M. Tanaka, Y. Abe, S. Furukado, K. Miwa, et al., Chronic kidney disease and carotid atherosclerosis, J. Stroke Cerebrovasc. Dis. 21 (1) (2012) 47–51. [18] D. Somnath, P. Jayantha, G.M. Kanti, A.D. Kumar, et al., Carotid artery intimamedia thickness in chronic renal failure patients with diabetes and without diabetes, Int. J. Med. Sci. 1 (3) (2012) 65–70. [19] J. Muscheites, A.A. Meyer, E. Drueckler, et al., Assessment of the cardiovascular system in pediatric chronic kidney disease: a pilot study, Pediatr. Nephrol. 23 (2008) 2233–2239. [20] S. Brzosko, U. Lebkowska, J. Malyszko, T. Hryszko, K. Krauze-Brzosko, M. Mysliwiec, Intima media thickness of common carotid arteries is associated with traditional risk factors and presence of ischemic heart disease in hemodialysis patients, Physiol. Res. 54 (2005) 497–504. [21] A. Philips, Medical System Company, Comparison of Image Clarity, SonoCT Real-Time Compound Imaging Versus Conventional 2D Ultrasound Imaging, ATL Ultrasound, Report, 2001. [22] T. Elatrozy, A.N. Nicolaides, T. Tegos, A. Zarka, M. Griffin, M. Sabetai, The effect of B-mode ultrasonic image standardization of the echodensity of symptomatic and asymptomatic carotid bifurcation plaque, Int. Angiol. 17 (3) (1998) 179–186. [23] J.S. Lee, Refined filtering of image noise using local statistics, Comput. Graph. Image Process. 15 (1981) 380–389. [25] I. Wendelhag, Q. Liang, T. Gustavsson, J. Wikstrand, A new automated computerized analyzing system simplifies reading and reduces the variability in ultrasound measurement of intima media thickness, Stroke 28 (1997) 2195–2200. [26] D.J. Williams, M. Shah, A fast algorithm for active contours and curvature estimation, Int. J. Graph. Vis. Image Process. Image Underst. 55 (1992) 14–26. [27] J.M. Bland, D.G. Altman, Statistical methods for assessing agreement between two methods of clinical measurement, Lancet 1 (8476) (1986) 307–310. [28] W.J. Conover, Practical Nonparametric Statistics, 3rd ed., John Wiley & Sons, New York, 1999. [29] N. Ishizaka, Y. Ishizaka, E. Toda, K. Koike, et al., Association between chronic kidney disease and carotid intima-media thickness in individuals with hypertension and impaired glucose metabolism, Hypertens. Res. 30 (11) (2007) 1035–1041. [30] J. Paul, S. Dasgupta, K.M. Ghosh, Carotid artery IMT as a surrogate marker of atherosclerosis in patients with chronic renal failure on hemodialysis, N.Am. J. Med. Sci. 4 (2) (2012) 77–80. [31] A.N. Nicolaides, M. Sabetai, S.K. Kakkos, S. Dhanjil, T. Tegos, J.M. Stevens, The asymptomatic carotid stenosis and risk of stroke study, Int. Angiol. 22 (3) (2003) 263–272. [32] C.P. Loizou, M. Pantziaris, M.S. Pattichis, E. Kyriakou, C.S. Pattichis, Ultrasound image texture analysis of the intima and media layers of the common carotid artery and its correlation with age and gender, Comput. Med. Imag. Graph. 33 (4) (2009) 317–324. [33] F. Destrempes, J. Meunier, M.-F. Giroux, G. Soulez, G. Cloutier, Segmentation of plaques in sequences of ultrasonic B-mode images of carotid based on motion estimation and a Bayesian model, IEEE Trans. Biomed. Eng. 58 (8) (2011) 2202–2211.

C.P. Loizou et al. / Computers in Biology and Medicine 53 (2014) 220–229

[34] T. Drueke, V. Witko-Sarsat, Z. Massy, B. Descamps-Latscha, A.P. Guerin, et al., Iron therapy, advanced oxidation protein products, and carotid artery intimamedia thickness in end-stage renal disease, Circulation 106 (2002) 2212–2217. [35] D. Saedi, M. Narooinejad, S. Saffari, M. Kkak, Comparison of intima-media thickness of the common carotid artery in dialysis and kidney transplant recipient patients, Exp. Clin. Transplant. 9 (1) (2011) 26–31. [36] R. Ekart, R. Hojs, T.H. Fabjan, B.P. Balon, Predictive value of carotid intima media thickness in hemodialysis patients, Artif. Organs 29 (8) (2005) 615–619. [38] G.M. London, A.P. Guerin, S.J. Marchais, B. Pannier, et al., Cardiac and arterial interactions in end-stage renal disease, Kidney Int. 50 (1996) 600–608. [39] S. Glagov, R. Vito, D.P. Giddens, C.K. Zarins, Micro-architecture and composition of artery walls: relationship to location, diameter and the distribution of mechanical stress, J. Hypertens. Suppl. 10 (6) (1992) S101–S104. [40] K. Jensen-Urstad, M. Jensen-Urstad, J. Johansson, Carotid artery correlates with risk factors for cardiovascular disease in a population of 55-year-old subjects, Stroke 30 (8) (1999) 1572–1576. [41] M.L. Yilmaz, A.R. Qureshi, J.J. Carrero, M. Saglam, et al., Predictors of carotid intima-media thickness in chronic kidney disease and kidney transplant patients without overt cardiovascular disease, Am. J. Nephrol. 31 (3) (2010) 214–221.

229

[42] G.A. Adeseun, D. Xie, X. Wang, M.M. Joffe, E.R. Mohler, Carotid plaque, carotid intima-media thickness, and coronary calcification equally discriminate prevalent cardiovascular disease in kidney disease, Am. J. Nephrol. 36 (4) (2012) 342–347. [43] C.P. Loizou, S. Petroudi, C.S. Pattichis, M. Pantziaris, A.N. Nicolaides, An integrated system for the segmentation of atherosclerotic carotid plaque in ultrasound video, IEEE Trans. Ultrasc. Ferroelectr. Freq. Contr. 61 (1) (2014) 86–101. [44] C.P. Loizou, M. Pantziaris, C.S. Pattichis, E. Kyriakou, M-mode state-based identification in ultrasound videos of the atherosclerotic carotid plaque, in: Proceedings of the 4th Intenational Symposium on Communications, Control and Signal Processing, ISCCSP, Limassol, Cyprus, March 3–5, 2010, p. 6. [45] C.P. Loizou, C. Theofanous, M. Pantziaris, T. Kasparis, Despeckle filtering software toolbox for ultrasound imaging of the common carotid artery, Comput. Methods Progr. Biomed. 114 (1) (2014) 109–124. [46] C.P. Loizou, C. Theofanous, M. Pantziaris, T. Kasparis, P. Christodoulides, A.N. Nicolaides, C.S. Pattichis, Despeckle filtering toolbox for medical ultrasound video (Special issue on Biomedical Monitoring Technologies), Int. J. Monitor. Surveill. Technol. Res. (IJMSTR): 4 (1) (2013) 61–79.

Manual and automated intima-media thickness and diameter measurements of the common carotid artery in patients with renal failure disease.

The objective of this study was to investigate differences in intima-media thickness (IMT) and diameter (D) measurements of the common carotid artery ...
2MB Sizes 3 Downloads 7 Views