Analysis of Multicontrast Carotid Plaque MR Imaging Huijun Chen, PhDa, Qiang Zhang, MSb, William Kerwin, PhDc,* KEYWORDS  MR imaging  Atherosclerosis  Morphology  Composition  Contrast enhancement  Biomechanics  Image processing

KEY POINTS  Quantitative analyses of atherosclerotic plaque are enabled by image processing.  Analyses of traditional MR images (T1-weighted, T2-weighted, and so forth) yield measurements of plaque morphology and composition.  Dynamic contrast-enhanced (DCE)–MR imaging and kinetic modeling can be used to assess plaque perfusion characteristics related to inflammation.  Coupling these techniques with computational models reveals plaque biomechanical forces.

In clinical practice, luminal stenosis measured from angiography is the most widely used parameter for risk prediction of atherosclerotic plaque.1 Together with the development of high-resolution multicontrast vessel wall MR imaging techniques, vessel wall features have drawn more attention in addition to luminal stenosis,2–6 including morphologic, compositional, physiologic, and hemodynamic features. Morphologic features of atherosclerotic plaque usually quantified from multicontrast carotid MR imaging are mainly used to evaluate the plaque burden, including vessel wall thickness, area, volume, and some normalized indexes, such as the normalized wall index (NWI).7–16 Segmentation of the lumen and outer-wall boundaries from blackblood carotid MR images is the key to acquiring

those quantitative morphologic features. Some automatic lumen/outer-wall boundary segmentation algorithms and how to measure those plaque burden features from the lumen/outer-wall boundaries are discussed. Atherosclerotic plaque has complex contents. Researchers have found that components have different signals in different weightings in multicontrast carotid MR images,12 providing a unique opportunity to identify those components. Many studies have found that some compositional features detected by multicontrast MR imaging can be used for vulnerable plaque identification, such as the thin fibrous cap, large necrotic core, and intraplaque hemorrhage.12,17–21 Methods of segmenting the plaque components, including multicontrast image registration and pattern recognition–based composition detection methods, are discussed.

Disclosure Statement: W. Kerwin is a former employee of VPDiagnostics, Inc (through 2013) and holds several patents related to the work. The remaining authors have nothing to disclose. a Department of Biomedical Engineering, Center for Biomedical Imaging Research, School of Medicine, Tsinghua University, Room No. 109, Haidian District, Beijing, China; b Department of Biomedical Engineering, Center for Biomedical Imaging Research, School of Medicine, Tsinghua University, Room No. 120, Haidian District, Beijing, China; c Department of Radiology, School of Medicine, University of Washington, 850 Republican Street, Seattle, WA 98109, USA * Corresponding author. E-mail address: [email protected] Neuroimag Clin N Am 26 (2016) 13–28 http://dx.doi.org/10.1016/j.nic.2015.09.002 1052-5149/16/$ – see front matter Ó 2016 Elsevier Inc. All rights reserved.

neuroimaging.theclinics.com

INTRODUCTION

14

Chen et al Recently, advances in carotid plaque MR imaging and analysis techniques have allowed researchers to evaluate atherosclerosis at the physiologic level. Physiologically, plaque vascularity and accumulations of macrophages are highly associated with plaque instability.22,23 To quantify the plaque angiogenesis and inflammation, DCE, an MR imaging technique, has been proposed and successfully validated.1,24,25 This article introduces the processing and modeling of DCE carotid plaque MR images. Lastly, hemodynamic conditions are believed related to pathogenesis of atherosclerotic plaque.26,27 Methods of extracting hemodynamic features from carotid plaque MR imaging, including the wall shear stress (WSS) and tensile stress, are discussed.28–34

image to image. One such technique is the Markov shape model (Fig. 1), in which the expected shape in the subsequent image is estimated based on the shape in the current image.36

3-D Boundary Detection As imaging methods have advanced to allow black-blood preparation in isotropic 3-D images, a new approach to boundary detection has emerged in which the entire lumen or outer-wall volume is segmented as a whole, rather than as a series of parallel contours. One such approach uses graph-cuts global optimization (Fig. 2) to simultaneously extract the lumen and vessel wall.37 Constraints ensure that the lumen is contained within the larger vessel wall volume.

Plaque Burden Quantification

PLAQUE BURDEN Plaque burden is a measure of atherosclerotic plaque size. Based on the imaging and processing protocol, plaque burden metrics can be 2-D crosssectional vessel wall thickness, area, 3-D volume, and other metrics. No matter which plaque burden metric is chosen, precise segmentation of the inner and outer boundaries of the vessel wall is an essential step. Black-blood carotid plaque MR imaging can show the vessel wall clearly, allowing automatic segmentation. Segmentation methods are in 2 categories: 2-D–based segmentation and 3-D–based segmentation. The 2-D segmentation methods detect the boundaries in crosssectional images, whereas the 3-D segmentation techniques segment the surface directly. After lumen and outer-wall boundaries are detected, a range of morphology metrics can be calculated. Imaging protocols For plaque burden assessment, the protocol should include: T1-weighted, 2-D or 3-D acquisitions, black-blood preparation, fat saturation, in-plane resolution less than 1 mm, slice thickness 2 mm or less.

2-D Boundary Detection Detection of the lumen and outer-wall boundaries has been accomplished most often by placing contours on stacks of 2-D cross-sectional images. Active contour methods, such as B-spline snakes, are available that deform an initial contour estimate to match the boundary apparent in the image.35 For greater automation, techniques have been developed that propagate a contour from

After the lumen and outer-wall boundaries have been identified, plaque burden metrics, such as thickness, area, and volume, can be calculated. Vessel wall thickness calculation Because of the various shapes of the diseased carotid artery, thickness measurement is a challenge. The key to calculating thickness is to find the matching lines connecting corresponding points between the lumen and outer-wall contours. Such correspondence, however, cannot be easily defined in carotid artery with severe plaque. The intuitive matching method of shortest distance may generate wrong matching lines, such as lines that cross the lumen (Fig. 3). Han and colleagues38 propose using the Delaunay triangulation technique to find the corresponding points, which can produce a stable result for various plaque shapes. Once the Delaunay triangulation is completed, the matching lines can be defined by the triangular midlines of all triangles. Vessel wall area and volume In addition to thickness, vessel wall area is a commonly used burden metric. The vessel wall area can be calculated by measuring the area enclosed by the outer-wall boundary minus the area enclosed by the lumen boundary. With known area, the plaque volume is calculated using the Simpson rule, which sums the areas of all images times the separation between images.39 Normalized wall index Although the absolute value of vessel wall area measurement can reflect the plaque burden, it does not account for the variance of carotid artery size in the population, which may introduce bias in clinical use. To solve this problem, a normalized metric has been proposed35: the NWI, which

Analysis of Carotid Plaque MR Imaging

Fig. 1. Image (A) is proximal to image (B) and is used to refine the search space of the Markov shape model (B). (C, D) Examples of initializations obtained, respectively, without and with knowledge of the preceding shape. (From Underhill H, Kerwin W. Markov shape models: object boundary identification in serial magnetic resonance images, in proceedings 14th scientific meeting. International Society for Magnetic Resonance in Medicine. vol. 14. 2006. p. 829.)

normalized the plaque area by the vessel size. NWI is defined as the area of the vessel wall divided by the area of the lumen plus wall and ranges from approximately 0.4 for a normal artery to near 1.0 for a highly stenotic artery. In some cases, NWI is calculate in terms of wall and lumen volumes instead of areas.14

Validation, Reproducibility, and Clinical Studies of the Morphologic Metrics Yuan and colleagues11 have shown that the maximal wall area measurements from in vivo and ex vivo MR imaging strongly agree. In another study comparing in vivo and ex vivo plaque MR images, the correlation coefficients were found as high as 0.92 for wall volume measurement, 0.91 for maximum wall area measurement, and 0.90 for minimum lumen area measurement.10 In a histology validated study,12 the area measurements of wall can be correlated with histology (r 5 0.84; P

Analysis of Multicontrast Carotid Plaque MR Imaging.

Plaque imaging by MR imaging provides a wealth of information on the characteristics of individual plaque that may reveal vulnerability to rupture, li...
566B Sizes 1 Downloads 12 Views