Computer-aided quantification of contrast agent spatial distribution within atherosclerotic plaque in contrast-enhanced ultrasound image sequences
Introduction
Contrast-enhanced ultrasound (CEUS) is a noninvasive imaging modality, applying microbubble-based contrast agents to traditional medical ultrasound to increase the backscatter of the ultrasound signal [1]. It enables detailed visualization of blood flow and perfusion and thus covers broad medical application.
In recent years, the use of CEUS continues to grow in the assessment of carotid atherosclerotic plaques [2], [3], [4], [5], [6]. Vulnerable plaques, also known as “rupture-prone plaques,” have a high probability of undergoing rapid progression and rupture, causing local thrombosis and embolism, and thus leading to stroke, one of the major causes of death worldwide [7]. Recent evidence has linked elevated plaque vulnerability with plaque angiogenesis and has demonstrated feasibility of CEUS as a noninvasive way for evaluation of intraplaque neovascularization (IPN) [5]. Therefore, CEUS could be of importance to assess plaque vulnerability and identify vascularized, “high-risk” lesions before rupture [8].
For assessment of plaque vulnerability, it is valuable to measure the presence and degree of IPN in CEUS images. The IPN is traditionally measured by applying a qualitative visual approach, usually by using a discrete grading system. Coli et al. [9] categorized the echogenicity as low if no bubbles were detected within plaques and as high if extensive contrast enhancement was depicted. Staub et al. [10] categorized the degree of IPN at CEUS as absent (grade 1), moderate (grade 2), or extensive (grade 3). However, these qualitative methods are subjective and affected by inter-observer and intra-observer variability [11], [12]. Thus, it is required to develop computer-aided methods for objective quantification of IPN.
There are two categories of computer-aided methods. One is intensity based, and the other is area based. Papaioannou et al. [13] measured the mean and median intensities of pixels in the plaque before and after injection of microbubbles. Huang et al. [14] and Xiong et al. [15] calculated average intensities in the plaque, including the baseline intensity, peak intensity, and enhanced intensity equaling to the peak intensity subtracting the baseline intensity. Furthermore, Xiong et al. [15] calculated the ratio of enhanced intensity in the plaque to that in the lumen. Slightly differently, Moguillansky et al. [3] used the ratio of peak intensity in the adventitia to that in the lumen. However, in all these intensity based studies, the intensity is averaged as the mean or median gray levels of pixels in a region of interest (ROI), so that the information of spatial distribution of contrast agents is neglected. Hoogi et al. [4] proposed an area based method by calculating the ratio of the IPN area to the total plaque area. This area ratio measures the relative extent of contrast agents inside a plaque, with a larger value representing more extensive spatial distribution of contrast agents. It is different from the intensity based quantities, as illustrated with an example in Fig. 1. The area ratio is proved to be correlated well with histopathologic results, demonstrating the great value of incorporating the information of spatial distribution in quantification of IPN on CEUS [4].
For the calculation of this area ratio, named the traditional area ratio (AR,H), CEUS images of a plaque are segmented offline frame by frame in a video sequence that is previously acquired in real time during patient examination. The image segmentation is used to detect IPN regions in each frame, and then an accumulation process is performed by integrating these regions from all frames to obtain the total area of IPN. However, there arise four questions.
First, CEUS images are corrupted by speckle noise, which makes robust image segmentation a difficult task [16]. It is complicated, cumbersome and time-consuming to segment a CEUS sequence frame by frame. When the accumulation of the segmented frames is conducted to derive the total IPN area, the segmentation error in all frames might also be accumulated. How can we determine the IPN area in a more accurate and efficient way?
Second, a CEUS sequence acquired from pre-injection of contrast agents to post-injection usually consists of hundreds and thousands of frames, so that a sub-sequence containing several cardiac cycles needs to be selected from the long sequence to simplify the computing [4], [17]. But for an effective and objective analysis, which portion of the long sequence should be selected?
Third, it needs electrocardiogram (ECG) gating for the selection of sub-sequence [4], [13]. In clinical CEUS examinations, we may not always record ECGs; even if we do, the recorded ECGs may not be synchronized with CEUS. In that case, can we retrieve cardiac cycles from the CEUS data themselves?
Finally and most importantly, there is detailed information of contrast agent spatial distribution that both the intensity and area based methods have ignored. Coli et al. [9] stated that the neovascularization should derive mainly from the adventitial vasa vasorum network and, therefore, progressively grow from the external layers toward the plaque core. In addition to the area ratio, the uniformity or nonuniformity of the contrast agents may also reflect the distribution characteristics of contrast agents [18]. Therefore, the degree of neovascularization invading to the plaque core and the uniformity of the contrast agents within the plaque appear to be promising features describing the spatial distribution of contrast agents. From the viewpoint of image processing, these features can be regarded as measures of texture within the plaque [19]. Now the question is: how can we quantify them?
The contribution of the paper is twofold. First, aiming at answering the aforesaid first three questions, we propose a computer-aided method for more objective and convenient detection of IPN. Second, in order to answer the last question, we propose extracting texture features to quantify the spatial distribution of contrast agents at IPN areas, including texture features based on grayscale images and binary images.
Section snippets
Materials and methods
An overview of the methods is summarized in Fig. 2 and the details are described in this section. All algorithms were implemented by MATLAB R2007a (The MathWorks, Natick, MA, USA).
Results
Experimental results are presented as follows.
Physical interpretation of texture features
Eight quantitative texture features belonging to two categories all exhibited significant difference between qualitative grades. The CDD, DD and RDD are the texture features derived from binary images. The CDD measures the degree of neovascularization noninvading to the plaque core, and the DD and RDD represent the spatial deviation and dispersion of the neovascularization. Therefore, a larger value of CDD and smaller values of DD and RDD indicate less extent of IPN, consistent with the
Conclusions
In conclusion, a computer-aided approach is proposed in this paper for objectively and conveniently quantifying the spatial distribution of contrast agents within plaques in CEUS image sequences. It consists of three algorithms, including cardiac cycle retrieval and sub-sequence selection, temporal mean image segmentation, and texture feature extraction. The results on 33 plaques demonstrated the usefulness of the method for objectively and conveniently quantifying the spatial distribution of
Acknowledgments
This work is supported by the Shanghai Natural Science Fund for Youth Scholars (No. 12ZR1444100), the National Natural Science Foundation of China (No. 61171032), and the Chenguang Project (No. 11CG45), the Scientific Innovation Fund (No. 12YZ026) and the Young Teacher Cultivation Project (No. shu11047) of Shanghai Educational Committee.
References (36)
- et al.
The humble bubble: contrast-enhanced ultrasound
Radiography
(2011) - et al.
Quantitative analysis of ultrasound contrast flow behavior in carotid plaque neovasculature
Ultrasound Med. Biol.
(2012) Contrast ultrasound imaging of the carotid artery vasa vasorum and atherosclerotic plaque neovascularization
J. Am. Coll. Cardiol.
(2006)- et al.
Contrast-enhanced ultrasound imaging detects intraplaque neovascularization in an experimental model of atherosclerosis
JACC Cardiovasc. Imaging
(2010) - et al.
Contrast-enhanced ultrasound imaging of intraplaque neovascularization in carotid arteries: correlation with histology and plaque echogenicity
J. Am. Coll. Cardiol.
(2008) - et al.
New quantification methods for carotid intra-plaque neovascularization using contrast-enhanced ultrasound
Ultrasound Med. Biol.
(2014) - et al.
Contrast-enhanced ultrasound imaging of carotid plaque neo-vascularization: accuracy of visual analysis
Ultrasound Med. Biol.
(2014) - et al.
In-vivo imaging of carotid plaque neoangiogenesis with contrast-enhanced harmonic ultrasound
Int. J. Cardiol.
(2009) - et al.
Speckle filtering of ultrasonic images using a modified non local-based algorithm
Biomed. Signal Process. Control
(2011) - et al.
Real-time texture analysis for identifying optimum microbubble concentration in 2-D ultrasonic particle image velocimetry
Ultrasound Med. Biol.
(2011)