Quantification of Atherosclerotic Plaque Elasticity Using Ultrasonic Texture Matching

The composition of an atherosclerotic plaque is a major determinant of its vulnerability, i.e. proneness to rupture. This paper proposes an ultrasonic texture matching method, which is shown to reflect the distribution of elastic modulus and is a potential method for quantitatively analyzing different plaque components based on B-mode cine-loops. We performed an in vitro study, employing plaque phantoms fabricated using polyvinyl alcohol. Firstly, the phantoms with two components (soft tissue: 60.9 ± 6.8 kPa; surrounding tissue: 248.8 ± 12.1 kPa) were fabricated. Soft tissue occupied 10%, 40% and 64% of the total plaque volume. Further, three tissue components (soft tissue: 60.9 ± 6.8 kPa; hard tissue: 248.8 ± 12.1 kPa; surrounding tissue: 310.3 ± 14.1 kPa) were made. Soft and hard tissues occupied 32% of total plaque volume, respectively. For our in vivo study, six mice with ApoE knockout and three New Zealand white rabbits with abdominal aortic balloon injury were evaluated. Ultrasound cine-loop data of plaques were collected to calculate elastic modulus, then the plaque tissues were removed for further histology examination. The cine-loop data in vitro study and in vivo study were acquired by an ultrasound micro-imaging system (VisualSonics Vevo2100). In the phantom experiment, the mean elastic moduli of two component phantoms were 60.4 ± 7.6 kPa (soft tissue) and 198.5 ± 12.5 kPa (surrounding tissue), respectively. Similarly, the mean elastic moduli of three component phantoms exhibited 90.2 ± 6.2 kPa (soft tissue), 184.3 ± 11.6 kPa (hard tissue) and 381.6 ± 3.8 kPa (surrounding tissue), respectively. In animal experiments, the percentage of lipid tissue and calcification regions was also quantified in mice and in rabbit experiment with the histological results. This suggests that the texture matching method may be a potential method to identify lipid component of plaque and to predict the vulnerability of atherosclerosis plaques noninvasively.

components, especially, soft tissue components (large lipid core) for predicting vulnerable plaque before rupture in a clinical setting.
Various diagnostic invasive methods have been investigated to identify plaque components. Intravascular ultrasound (IVUS) was proposed for distinguishing plaques morphology [7], [8]. Intravascular ultrasound elastography (IVUSE) is a technique based on IVUS [9], [10], which can be used to qualify or quantify specific histologic components of atherosclerotic plaques by measuring tissue elastic properties to identify vulnerable plaques [11]- [13]. Virtual-histology intravascular ultrasonography (VH-IVUS) [14], [15] was widely used to qualitatively identify coronary plaque specific components including fibrous tissue, fatty tissue, necrotic core and calcification [16], [17]. Optical coherence tomography (OCT) was widely used for coronary plaque characterization [18], [19], especially the detection of thin-cap fibro atheroma caused plaque rupture [20]- [22]. Moreover, OCT characterizations for various plaque components such as lipid content, fibrous cap thickness and calcification were also validated by histology [23], [24]. Computed tomography is also a technology for detecting the component of atherosclerotic plaques, but it also has limitation to evaluate necrotic core and underestimate calcification (Weert, et al. 2006, Obaid, et al. 2013. Though these techniques were capable of distinguishing the characteristics and components of the atherosclerotic plaques, their invasive nature presents a limitation. Magnetic Resonance Elastography (MRE) is a noninvasive technology for quantitatively assessing the mechanical properties of tissue, especially for application to estimate the stiffness of organs such as brain, breast, blood vessel, heart, kidney, lung, and skeletal muscle [25]- [28]. In addition, Magnetic Resonance Imaging (MRI) combined with T2 mapping can be used to discriminate tissue characteristics and plaque components with desired accuracy in comparison with histology [29]- [31]. However, MRI is very expensive for widespread use and also time consuming. Moreover, the capability of Positron Emission Tomography (PET) for quantification of plaque inflammation has been evaluated to predict the risk of atherosclerotic plaque rupture, but the disadvantage is that contrast agents are required and the human body is exposed to radiation [32], [33].
Ultrasound examination is a common and non-invasive technique for detecting the pathology of internal body structures such as tendons, muscles, joints, blood vessels, and internal organs. Recently, ultrasound techniques were developed to identify the vulnerability of the plaque with the advantage of being inexpensive and non-invasive. Noninvasive ultrasound techniques based on radiofrequency (RF) data were used to estimate the strain of plaques [34]- [36]. Roy Cardinal et al. reported that ultrasound noninvasive vascular elastography (NIVE) based on RF data is able to quantify axial strain, shear strain, and translation motion. Thirty-one patients were enrolled in the study, and the results showed that ratio of cumulated axial strain to cumulated axial translation of vulnerable plaque were apparently higher than nonvulnerable plaques without neovascularity [37].
Kanai et al. provided a noninvasive method based on envelope data for evaluating the regional elasticity of tissue surrounding atherosclerotic plaque. Nine iliac arteries with plaques were used in the experiment and the elasticity of plaques were shown to be about 81 ± 40 kPa and 1.0 ± 0.63 MPa, which were used as the reference parameters [38]. However, the limitation was that the amount of RF data was large and the process of calculation was time consuming. Widman et al. adopted a developed speckle tracking (ST) algorithm based on B-mode, which was used to assess plaque strain and was verified in phantoms and in vivo using sonomicrometry. For the in vitro and in vivo results, radial and longitudinal limits of agreement (LOA) were well correlated and occurred between ST and sonomicrometry peak strains [39]. Shear wave elastography (SWE) can also be a measure to quantify elastic modulus distribution in atherosclerotic plaque. However, the reproducibility, accuracy and resolution of the SWE technique needs to be further improved [40]- [43].
Vulnerable plaques usually have a large lipid core that is an important determinant for detection. The aim of our study is to use the proposed texture matching method to quantitatively analyze the elastic modulus distribution and identifies lipid components in plaques. The method is potentially a noninvasive method to evaluate the vulnerability of atherosclerotic plaque based on B-mode images. The B-mode data can be acquired easily in clinical ultrasound equipment and is inexpensive and non-invasive. The feasibility of this method was demonstrated using in vitro plaque phantoms made from polyvinyl alcohol (PVA) cryogel and in vivo animal abdominal aortic plaques.

A. THE TEXTURE MATCHING METHOD
Our study provides a texture matching method to calculate the distribution of elastic distribution of plaques. The procedures of the method are as follows: firstly, f (1) are N frames of B-mode images extracted from a cine-loop, where (x, y) corresponds to the coordinates of a pixel in the image plane. Then, g (n) (x, y) and g (n+1) (x, y) are selected from two successive images f (n) (x, y) and f (n+1) (x, y) (1 < n < N ) are defined as the region-of-interest (ROI), and ROI is a rectangular area that contains the entire plaque. The texture matching method starts when n = 1 and repeats with the same size ROI translated exactly the same distance as the estimated displacement, until n = N − 1. Here, the RIO is the target region for calculating the elastic modulus distribution. Each ROI is divided into a grid of small sections known as interrogation windows. After a series of interrogation windows are divided, the most similar texture distribution of interrogation windows of two consecutive B-mode images are matched. The texture displacement of each interrogation VOLUME 8, 2020 window is obtained via a normalized two-dimensional crosscorrelation algorithm with sub-pixel method, filter and interpolation method, and these methods have been explained in our previous studies [44], [45].
For improving the accuracy of displacement, the Gaussian peak fitting is used for the sub-pixel analysis. Then, in the displacement field, the vector will be erased if it is larger than the median of all the nine vectors plus the threshold (1.7-3.0) times the standard deviation of all the vectors, or if it is smaller than the median minus the threshold times the standard deviation when each vector is compared to the eight vectors surrounding it.
The filtering is used to modify the non-uniform vectors. The bilinear interpolation method is used to supplement the missing vector, and the rotation and deformation of ROIs in the images are calculated by the gradient of the two-dimensional translational displacement described above. Two-dimensional displacement ν(x, y, n) of the geometric transformation is acquired, where x and y are image coordinates, and n is the frames of B-mode images.
A previous detailed description of the algorithms has been reported by our previous studies [44], [45]. All the algorithms are implemented in Matlab (64-bit, R2016b, MathWorks Inc., Natick, MA, USA).

B. ELASTIC MODULUS EVALUATION
After the estimated displacements ν(x, y, n) are acquired, the displacement gradient ε is detected by each axial layer texture displacement of each plaque. The displacement gradient (strain) of each layer with a constant thickness of h 0 is obtained as follows: ε(x, y, n) = (ν(x + 1, y, n) − ν(x, y, n))/h 0 . (1) The maximum strain of each layer during one cardiac cycle is obtained by: Then the elastic modulus distribution is calculated. Phantom and animal plaques are assumed to be incompressible, the elastic modulus distribution is estimated by Hasegawa [46]: where M and R il are the number of layers and the inner radius of the l-th layer, respectively, and P is the pulse pressure difference. The layers indicate that the plaque has been divided n layers, and axial displacement is calculated each layer in the ROI region [44]. Here, h 0 is the thickness of the axial adjacent layer blood vessel wall in one cardiac cycle and ε max is the largest displacement gradient for each layer in one cardiac cycle. The assumptions are that the isotropy and Poisson ratio are 0.5. In previous studies, the axial strain or lateral strain calculated assumes that the entire vessel wall is homogeneous (only one value is calculated). However, the elastic modulus of each layer of tissue in the plaque is different. Our method potentially has the capacity for calculating the strain of each layer of tissue and then estimating the elastic modulus distribution.

C. PHANTOM PREPARATION
Polyvinyl Alcohol Cryogel (PVA) (MW 146000-186000; 99+% hydrolysed, Sigma-Aldrich, USA) is widely applied to phantom experiment and is adopted to mimic vessel and plaque phantom [47]- [49]. In this study, a PVA phantom consists of 87% ultrapure water, 10% PVA powder and 3% cellulose powder (Type 20, Sigma-Aldrich, USA) three components. The production procedure is as follows: firstly, stir PVA powder in ultrapure water continually using a DF-101S magnetic stirrer (Gongyi Yuhua Instrument Co., Ltd., Gongyi, Heinan, China) at 20 • C for 1.5 hour. Then, add the cellulose powder and the stir them at a 95 • C for 1 hour. After that, the solution is cooled to 20 • C and poured into a metal mould. As shown in Figure 1, the metal mould consists of a metal shell, a metal bar and a metal cap composed of two identical parts. Finally, the formation of the cryogel phantoms are induced by placing the model into a freezer and subjected to several repeated freeze-thaw cycles.
As a cryogel, the material obtains its rigidity via a freezethaw process and the stiffness of the material increases as the number of freeze-thaw cycles increase. Each cycle consists of two phases freezing at −20 • C for 12 hours and thawing at 20 • C for 12 hours. In Figure 2, all the samples experienced 1-8 freeze-thaw cycles, and elastic modulus of the phantoms are between 50 kPa-350 kPa.
Four different phantom sets were fabricated in this study: (1) Three phantoms with two components: soft tissue that underwent one freeze-thaw cycle PVA accounting for 10%, 40% and 64% of the total plaque volume, respectively, and surrounding tissue of the plaque phantom that underwent five freeze-thaw cycles. (2) A phantom with three components: soft tissue that underwent one freeze-thaw cycle and hard tissue that underwent three freeze-thaw cycles accounting for 32% respectively of the total plaque volume, the surrounding tissue of phantom plaque underwent eight freeze-thaw cycles  (plaque: length: 15 mm; degree of stenosis: 50%; phantom: wall thickness: 1 mm; inner radius: 5 mm; length: 100 mm).

D. PHANTOM TENSILE EXPERIMENT
The average elastic moduli of PVA cylindrical samples in different freeze-thaw cycles were verified by an electronic universal material testing machine (CMT-6104, New Sans Machinery Co., Ltd., Shenzhen, Guangdong, China). The methods are as follows: (1) measure the height and diameter of each sample by vernier caliper, mean values are obtained by five times; (2) obtain the force F exerted on the samples and height change L of the samples by pressure sensor and displacement sensor. The elastic moduli are calculated by: where σ and ε are the stress and strain exerted on the samples, d and L are diameter and initial height of each samples. Moreover, in Figure 2, all the samples experienced 1−8 freeze-thaw cycles, and elastic modulus of the phantoms are between 50 kPa −350 kPa.

E. IN VITRO PLAQUE PHANTOMS EXPERIMENTS
The diagram of experimental system is shown in Figure 3. A pulsatile pump (Model 55-3305; Harvard Apparatus, Holliston, MA, USA) was used to mimic heart to provide power for human blood flow. The heart rate, pulsating flow and systolic-diastolic ratio were 45 beats min-1, 15L/stroke and 35/65, respectively. Systolic-diastolic ratio is the ratio of blood flow per minute during systolic and diastolic period. A buffer device was served as reducing the strong impact from pulsatile pump. A pressure transducer (HDP708, HeDi Sensor Instrument Co., Ltd., Foshan, Guangdong, China) connected with a digital oscilloscope (Teledyne LeCroy, Wavesurfer 3024, Chestnut Ridge, NY, USA) was installed between the buffer device and phantoms. The oscilloscope recorded the intraluminal pressure logged by the pressure transducer. The cine-loops of plaque were captured by MS550S transducer (transmit frequency: 40 MHz; frame rate: 117 Hz) and Vevo 2100 high resolution ultrasound imaging system (Vevo 2100, VisualSonics, Inc., Toronto, Canada). Two different ultrasound transducers are used in the paper because the depth of plaques are different in the in vitro and in vivo experiments. Note that B-mode images were acquired for a 4-5 s cine-loop and were decomposed into 350-450 frame images for each phantom. The focal depth was 6 mm and field of view was 15.5 mm (depth) by 10 mm (width). The size of interrogation window was 285 × 103 (pixels2) and the overlap was 55%. The cineloop of each phantom was collected for three times. The area of different elastic modulus of soft tissue, hard tissue and surrounding tissue area calculated by the texture matching method were quantified by manual segmentation with the help from a doctor using ImageJ Software (64-bit, version 1.47v, National Institutes of Health, Bethesda, MD, USA). The percentage of these tissues calculated by the texture matching method were compared to the phantoms fabricated above.

F. IN ANIMAL EXPERIMENTS
The texture matching method was investigated with an in vivo preliminary study performed with six male ApoE knockout mice and three adult male New Zealand white rabbits. And all the mice (weight: about 40g) were fed a high-fat diet containing 21% lard and 0.15% cholesterol for a period of 5 to 20 weeks. After the plaques were found, the ultrasound experiment was performed about the 20th week and the mice were subsequently euthanized. Each rabbit model of an abdominal aortic plaque was established by under general anesthesia and underwent balloon-induced abdominal aortic endothelial damage. Following balloon-induced damage, both the rabbits were maintained on a high-fat diet (120-140 g/day of 1% cholesterol and 99% standard rabbit diet) for a period of 5 to 24 weeks. The ultrasound experiment was performed on about the 24th week and the rabbits were subsequently euthanized. All the mice (20 weeks) and rabbits (about 24 weeks) were raised in individual cages in Peking University Shenzhen Hospital. The protocol was approved by the institutional review board of Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences.Mice experiment procedures were as follows: firstly, the mice were anesthetized with Avertin, and then abdominal VOLUME 8, 2020 regions were shaved. The B-mode movies of the abdominal aorta plaques were acquired over four cardiac cycles about 3-4s cine-loops were decomposed into 450-600 B-mode images, using a MS250 linear array transducer (transmit frequency: 20 MHz; frame rate: 100 Hz). The focus depth range was 2-3 mm depending on the individual differences. Blood pressure was referred to the standard value of mice blood pressure. In this study, a final 139×30 (pixels 2 ) interrogation window with an overlap ratio of 50% was used.
The rabbit experiment steps were as follows: firstly, rabbits were anesthetized with an intravenous injection of 3% Phenobarbital and fixed on the inspection table, then the abdominal areas were shaved. Next, the B-mode movies of the abdominal aorta plaques were acquired over four cardiac cycles about 3-4s cine-loop and 350-450 B-mode images, using a MS250 linear array transducer (transmit frequency: 20 MHz; frame rate: 100 Hz). The focus depth range was 5-5.5 mm depending on the individual differences. Blood pressure was referred to the standard value of rabbit blood pressure. In this study, a final 278 × 58 (pixels 2 ) interrogation window with an overlap ratio of 55% was used in texture matching analysis.
Vulnerability-Index (VI) was used to evaluate the overall degree of vulnerability [50], [51]. VI was calculated by the relation between analyzed unstable (U) and stable (S) features of the plaques. The formula for VI was expressed as VI i = U i /V i , where U includes unstable region, the lipid region area, and V includes stable region, the sum of fiber tissue and calcification area in the study, where i represents each mouse and rabbit.
All data were calculated three times and expressed as mean ± SD. Statistical significance was performed by paredsamples t-test, which were conducted using SPSS statistics (Version 20, IBM Corporation, Chicago, IL, USA). Statistical power (1−β) were performed by G * Power 3 (Version 3.1.9.4, Germany). Because of the small size in vivo experiment, statistical significance was performed by paired-samples t-test between elastography and histography.
Initially, ten ApoE knockout mice and five adult male New Zealand white rabbits were raised for the experiment. Because the integrated plaque tissue was hard to acquire, five mice and two rabbits expired in the process of animal raising. Also, the plaques did not form in six mice and five rabbits. Consequently, only six results from mice and three results from rabbits were adopted. Mouse and rabbit husbandry and all the procedures used in the study were performed in accordance with the guidelines and regulations of the Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences.

G. HISTOLOGY
The histology method of mice and rabbit is identical and the procedures are as follows: After the mice and rabbits were euthanized, the plaques were collected and were snapfrozen. A series of frozen sections (20 µm) were obtained for histological examination. The frozen sections were acquired  Figure 4 indicates the elastic modulus distribution of soft tissue respectively accounted for 10% (a), 40% (c) and 64% (e) of the total plaque volume of two component phantom plaques. It shows that the soft tissue (red region) embedded in the plaque can be clearly distinguished from surrounding tissues (blue region) by the elastic distribution. Additionally, the mean elastic modulus of the soft tissue is significantly lower than the elastic modulus within the surrounding tissue. The mean elastic moduli of three experiments in soft tissue are 43.7 ± 5.5 kPa, 67.2 ± 10.5 kPa and 58.2 ± 7.8 kPa, respectively, while the mean elastic moduli of surrounding regions are 195.0 ± 16 kPa, 199.8 ± 7.7 kPa and 200.5 ± 7.7 kPa, respectively. The mean elastic moduli are also compared to the experimental value verified on the electronic universal material testing machine in Figure 3. The mean elastic modulus of soft tissue subjected one freeze-thaw cycle is 60.9 ± 6.8 kPa, and the mean elastic modulus of stiff tissue experienced five freeze-thaw cycles is 248.8 ± 12.1 kPa. In comparison, a difference of 5.1% and 20.3% are compared between calculated mean values by the texture matching method and experimental value by tensile experiment in one and five freeze-thaw cycles. A three-component phantom is also measured by the texture matching method as shown in Figure 5. Three regions respectively experienced one, five and eight freeze-thaw cycles and the mean elastic modulus of eight freeze-thaw cycles is 381.6 ± 3.8 kPa calculated by texture matching method and mean elastic modulus of PVA subjected eight freeze-thaw cycles is 310.3 ± 14.1 kPa tested by the experiment. A difference of 23.0% is compared by the electronic universal material testing machine again.

B. ANIMAL EXPERIMENT
In order to ensure the reliability of the experimental results, only large and intact plaques were adopted as follows: six abdominal aortic plaque of mice (Plaque1, Plaque2, Plaque3, Plaque4, Plaque5 and Plaque6) and three abdominal aortic plaque of rabbits (Plaque1, Plaque2 and Plaque3) are evaluated in this section. Figure 6 (a) shows an example of a plaque (height: 0.6 mm, length: 3.9 mm) containing heterogeneous tissue in one abdominal aorta of a mouse. Based on the B-mode image, the plaque contains low echo intensity in the right region and high echo intensity in the left region. The elastic distribution of the plaque calculated by the texture matching method is displayed in Figure 6 (b). It shows that the elastic modulus of the tissue in the right region is apparently higher than the left. Figure 6 (c) shows hematoxylin and eosin histological staining that displays the morphological characteristic of the plaque and the outline of the plaque labeled by green line. Figure 6 (d) demonstrates Masson histological staining with a mass of dark blue region, which identifies the presence of collagen in the fibrous tissue. Figure 6 (e) shows Oil Red O staining, and red area is showed on the right side of the plaque that demonstrates the existence of lipid tissue. Figure 6 (f) shows that the plaque is absence of calcification verified by Von Kossa staining. The soft region and stiff regions and lipid tissue and calcification are quantified by ImageJ software, respectively and the percentage are showed in Table 1. Plaque 1 (the plaque in Figure 6) shows that lipid and calcification regions occupy 5.00 ± 0.44% and 5.76 ± 0.35% via elastography. Plaque 2 shows that the lipid and calcification regions occupy 34.34 ± 1.52% and 3.04 ± 0.44% via elastography, and 34.18 ± 0.56% and 4.30 ± 0.05% via histology. The percentage of lipid tissue and calcification calculated in plaque 1 and plaque 2 in Table 1 are in good agreement with histological examination. Similarly, other plaques (plaque 3, plaque 4, plaque 5, plaque 6) from mice also show great agreement in elastography and histological examination (lipid: n = 6, p = 0.987, power (1 − β) = 0.887). The VI of plaques of all mice have been calculated and supplemented in Table 1. Apparently, plaque 2 and plaque 3 contained more lipid region means high vulnerability. Figure 7 shows a rabbit example of composited components of abdominal aortic plaque. Figure 7 (a) shows the B-mode image of the plaque contained partial calcification (height: 1.8 mm, length: 3.0 mm). Figure 7 Table 2. Plaque 2 in Table 2 (the plaque in Figure 7) shows that  lipid and calcification region occupy 20.05 ± 1.90% and 7.51 ± 0.56% via elastography, and 15.56 ± 1.40% and 10.44 ± 0.72% via histology. The percentage of lipid tissue and calcification calculated from all the plaques from rabbits are in good agreement with histological examination. Similarly, other plaques (plaque 1 and plaque 3) from rabbits also show great agreement in elastography and histological examination in Table 2 (lipid: n = 3, p = 0.077, power (1 − β) = 0.697). The VI value of plaques of all rabbits have been calculated and supplemented in Table 2. Apparently, plaque 1 and plaque 3 contained more lipid region means high vulnerability.

IV. DISCUSSION
The composition of atherosclerotic plaque is a significant marker of the vulnerability of plaques [52], [53]. We propose a texture matching method to distinguish plaques with two and three components in phantom study. Further, the effectiveness of the method for evaluating elastic moduli distribution of animal plaques is validated by histological staining. The results demonstrate the potential of the texture matching method in estimating the elastic distribution of the plaques, which is promising in assessing the vulnerability of the atherosclerotic plaque and predicting the risk of cardiovascular disease.
The soft tissue area of elastography of plaque phantoms calculated by texture matching method was also quantified by an ImageJ Software. The percentage of area of one cycle region are 12.3%, 36.7% and 58.7% calculated by texture matching method, which are existed 2.3%, 3.3% and 5.3% difference with 3 phantom plaques, respectively. Furthermore, in the three components phantom, the percentage of area of one cycle and five cycles region were 26.4% and 27.7% percentage of the whole phantom plaque calculated by texture matching method, in comparison, and existed 5.6% and 4.3% difference with the phantom plaque. The cause of the deviation may be determined by problem of the experimental system, for example, the sealing of the experimental system may all result in some experimental errors. All of the factors could be related to low accuracy of the method. Moreover, selection of diagnostic window plays an important role in the accuracy and reliability of elastic modulus calculations in two consecutive B-mode images. If the diagnostic window is not large enough, it is difficult to implement the method because of insufficient plaque texture information in ROI. On the contrary, if the diagnostic window is too large, the spatial resolution will be affected. Normally, the selection of the interrogation window follows the 'onequarter rule' provided by Adrian [54], which means that the displacement of texture must less than or equal to one quarter the size of the interrogation window. In Figure 5-7, the boundary of the plaques looks uniform, sharp and clear, and it is because the sub-pixel method, the Gaussian peak filter and interpolation method are utilized in the acquisition of the elastic distribution.
The range of estimated elastic modulus in vitro experiment varies from 50 kPa to 350 kPa approximately, which is tested by the phantom tensile experiment (see Figure 2). This scope is basically consistent with the elasticity of biological tissues. Elastic moduli have been reported to vary in such a range for human tissues, especially for undiseased tissue, common carotid artery [55]. Further, the average elastic moduli of PVA phantoms in different freeze-thaw cycles were compared other identical materials studied in phantom experiment. Ramnarine et al. have found the mean elastic modulus of PVA phantoms subjected to two and five freeze-thaw cycles were 43 kPa and 170 kPa, respectively [40]. The elastic modulus is basically consistent with 56.4 kPa and 198.4 kPa calculated by our method (see Figure 3). While in Fromageau et al. study, PVA phantoms were 25 ± 3.0 kPa, 302 ± 35 kPa and 465 ± 53 kPa in one, five and eight freeze-thaw cycles measured by tensile experiment [56], were a slightly higher than our results. The different elastic modulus of PVA phantoms probably caused by the concentration of acoustic scatterer, the temperature and humidity of the freeze-thaw process and different algorithms.
In vivo study, only soft tissue and hard tissue components were distinguishable via the texture matching method. Lipid and fiber tissue were evaluated by histological verification, but intraplaque hemorrhage (IPH) and vasa vasorum in plaques could not be observed by our method. Notably, vasa vasorum is of significance as an indicator of vulnerability of plaque [57]. Further, the thickness of the fibrous cap was also referred to estimate the vulnerability of the plaque [52], whereas it was difficult to identify the specific thickness of fibrous cap by our method due to the resolution. The problem may possibly be resolved by acquiring a B-mode image of high resolution and applying a smaller interrogation window in the future. So, the method might be utilized in clinical ultrasound equipment to detect atherosclerotic disease in the near future.
We have also compared to previous methods, like ultrasound radiofrequency (RF) data and RF signal envelope data was used to estimate strain of plaques, while it was hard to acquire clinical data of raw clinical data and the quantity of the data of one case was too large, wasting a lot of processing time [37], [38]. Shear wave elastography was a common measure to quantify elastic modulus distribution in atherosclerotic plaque. However, the pulsation of the vessel wall is a limitation affecting the accuracy of the SWE and the method also based on assumptions of linear behavior in an isotropic, semi-infinite medium. Therefore, the method based by B-mode data is a more appropriate choice.

V. LIMITATIONS OF THE STUDY
Our study also has some limitations that exist in the experiment and need to be addressed in the future. Firstly, the plaque phantoms cannot be fabricated the same size as the real plaque because of the characteristics and difficulties of the plaque fabrication process. Besides, the frame rate (at least 80 Hz) of the ultrasound equipment is a key parameter in the experiment, because this parameter directly affects the feasibility of the cross-correlation algorithm. At present, most ultrasound imaging systems for small animals, even clinical equipment can provide higher frame rates. And eliminating the tissue motion is also an issue. In the experiment, all animals were anesthetized and maintained breathing while capturing the B-mode data to eliminate the affect of motion. Another limitation is that the elastic moduli of animal plaques in vivo cannot be calculated and verified. However, the absolute elastic modulus is not so crucial for the stiffness distribution and it is sufficient to determine in affirming the presence and percentage of lipid tissue. In addition, the blood pressure of rabbits is estimated by referring to the normal value. Moreover, we do not measure the blood pressure from mice and rabbits and the open question is to further verify our texture matching method by clinical experiment. Pressure guide wire can be a method to measure pressure in animal organs. In addition, in a clinical scenario, brachial artery pressure can be readily measured. The absolute values are not acquired from the in vivo experiment. The elastography combined with echo intensity is a reliable means for identifying the components of the plaque and the absolute values from the in vivo experiment can be the focus of future. Manually segmenting the plaque in ultrasound images is prone to deviation and in our experiment, the plaque profiles were extracted by a specialized physician. Animal modeling is also a limitation. Atherosclerosis may cause myocardial infarction, cerebral infarction and cerebral hemorrhage, leading to high mortality in the rabbit and mouse models. The animal validation study is relatively small, and the number of animal experiments and clinical experiments should be increased in the future.

VI. CONCLUSION
In conclusion, the study provided phantom verification and histological validation of the texture matching method for differentiating soft and stiff plaque tissues. The method quantified elastic modulus distribution to assess the vulnerability of the plaque. This technique may be of value in patients to prevent of the occurrence of stroke and other cardiovascular diseases and evaluate the effect of drug treatment for vulnerable plaque in a clinical setting. Since 2007, she has been with the Department of Ultrasound, Third Affiliated Hospital, Sun Yat-sen University, for more than ten years. Her research interests include the clinical application of the new technology of ultrasound, including application of contrast-enhanced ultrasound, ultrasound of liver transplantation, evaluation of vascular wall and plaque by elastography, and early diagnosis of atherosclerosis. She has authored or coauthored more than 30 refereed research articles related to above research fields. She is principal investigator of two projects, including the National Natural Science Foundation Grant of China. LILI NIU (Member, IEEE) received the master's degree in biomedical engineering from Northeast University, China, in 2009, and the Ph.D. degree in computer application technology from the Chinese Academy of Sciences (CAS), in 2012. She is currently an Associate Professor with the Shenzhen Institute of Advanced Technology, CAS. Her main research interests include developing advanced biomedical ultrasound techniques for measuring vascular biomechanics, and ultrasound neuromodulation. She has published more than 30 peer-reviewed journals in the biomedical engineering field, and applied more than ten patents. She is a principal investigator of six projects, including the National Natural Science Foundation Grant of China, and so on. VOLUME 8, 2020