Pixel Distribution Analysis to Identify Unstable Carotid Plaques
Is carotid plaque morphology ready for prime time?
Stroke rates achieved after carotid endarterectomy (CEA) for high-grade carotid stenosis (³60%) are significantly lower than those observed with best medical therapy alone. However, in the VA Asymptomatic1 and Asymptomatic Carotid Atherosclerosis Study,2 a majority (81% and 89%, respectively) of medically treated patients also remained stroke-free during follow-up. Increasing degrees of stenosis have not been associated with a correspondingly increased risk for stroke in asymptomatic patients, and there is continuing reluctance at several centers to revascularize asymptomatic patients with moderate carotid stenosis (60% to 79%). It is therefore important to identify additional markers to assess the risk of stoke and to identify patients who would benefit most from revascularization.
HISTOLOGIC EVIDENCE OF UNSTABLE CAROTID PLAQUES
Atherosclerotic plaques originate from fatty streaks which, over time, coalesce into a lipid core. Fibroatheromas form as fibrous tissue accumulates over the core and forms a fibrous cap. The notion of an unstable or vulnerable atherosclerotic plaque was first proposed for the coronary artery based on the observation that all coronary culprit lesions were not necessarily high-grade stenoses. Emerging information indicates that fibroatheromas are rendered unstable or vulnerable through an enlarging lipid core, intraplaque hemorrhage, fibrous cap thinning or rupture, and ulceration. These histomorphologic changes are associated with atheroembolization and have been observed in CEA specimens obtained from patients with symptomatic (transient ischemic attacks or stroke) carotid plaques.3,4 Large juxtaluminal lipid/necrotic cores may also be vulnerable to endovascular catheters, guidewires, balloons, or stent manipulation, and may predispose to atheroembolization during carotid artery stenting (CAS). Therefore, noninvasive plaque characterization would be extremely useful in several aspects of the management of carotid occlusive disease. It could help in stratifying asymptomatic patients for treatment; serial assessments could track responses to potentially "plaque-stabilizing" pharmacotherapy; it may also guide appropriate stent designs (eg, open- vs closed-cell). The challenge is to develop imaging techniques that can reveal this level of structural detail in the plaques of patients.
NONINVASIVE IMAGING FOR THE UNSTABLE PLAQUE
Currently available imaging modalities for assessing carotid plaque morphology (tissue composition and architecture) include MRI, CT, and ultrasound (US). MR images with a high signal-to-noise ratio are best obtained by magnetic coils positioned on the neck or within the carotid artery. Artifacts from carotid pulsations and respirations must be accounted for, and the cost of the multiple protocols required is high. Arterial wall assessment with CT is limited by a lack of tissue definition, although estimates of luminal stenoses are continuing to improve. Duplex US technology is less expensive, can be performed quickly and conveniently, and is ideally suited for serial assessments. Importantly, duplex US is already an integral part of the carotid evaluation for degree of stenosis using Doppler measurements. Concomitantly acquired B-mode images have traditionally not been utilized for more than a subjective assessment of plaque location and appearance. However, the quality of B-mode images has improved significantly and high-frequency transducers are now capable of high-resolution images at controlled angles and depths.
HISTORICAL PERSPECTIVE ON B-MODE IMAGING OF THE CAROTID PLAQUE
It was first suspected more than 2 decades ago that the echo patterns in B-mode images of carotid plaques could relate to tissue composition. Plaque echogenicity was qualitatively defined as the degree of acoustic brightness, and dense fibrocalcific lesions were noted to appear hyperechoic (bright) on visual examination of B-mode images. Gray-Weale5 used these observations to classify carotid plaques into pure hypoechoic (type 1), hypoechoic with small hyperechoic areas (type 2), hyperechoic with small hypoechoic areas (type 3), and pure hyperechoic (type 4). Symptomatic patients were noted to have predominantly types 1 and 2 lesions. However, variability in image acquisition and subjective quantification of echogenicity limited the accuracy of these assessments. El Barghouty6 added objectivity to the assessment of plaque echogenicity by measuring pixel brightness in the grayscale images. Low intensities (expressed as grayscale mean or median values) were observed more frequently in symptomatic plaques. However, because these were composite measures of the entire plaque, segmental areas of plaque instability were missed. This would explain why the composite grayscale cutoff value differentiating symptomatic from asymptomatic plaques varied in different reports. Geroulakos7 incorporated image variability into the Gray-Weale classification and suggested that heterogeneous plaques presented the highest risk for neurologic complications. This provided a general index of variation in appearance of the plaque. However, it is now becoming evident that quantification of specific plaque histologic features is the most critical issue in assessing risk for atheroembolization.
ACCURATE IDENTIFICATION OF TISSUE COMPOSITION AND ARCHITECTURE
The hallmarks of an ideal noninvasive assessment of morphology include an ability to quantify critical plaque histologic components (composition) and their distribution (architecture). Additionally, these measurements must be validated with anatomical comparisons, and the technique must demonstrate low observer variability. Different tissues reflect US differentially, resulting in B-mode images with variable brightness or pixel intensities. Using this principle, we developed a digital image-analysis algorithm (pixel distribution analysis [PDA]) that combines standardized US image acquisition, image normalization, and digital pixel segmentation with tissue mapping to identify plaque morphologic features.3 Longitudinal plaque images were obtained with a Sequoia 512 US machine (Acuson, Mountain View, CA) using a 7-13ÐMHz linear array transducer. The process was standardized to avoid variability in image acquisition (dynamic range 60 dB, preprocessing 0, persistence 2, postprocessing 0, gain -5 dB, and depth gain compensation linear). The images were further normalized by linear scaling, using a commercially available image-editing software (Adobe Photoshop 6.0, Adobe Systems, San Jose, CA). These steps allowed valid comparisons to be made between patients.
We first measured how echo intensities of B-mode images varied with the type of tissue being interrogated. These observations form the basis for PDA. The major histologic features of an unstable plaque include a thin fibrous cap, large and superficially located lipid core, and intraplaque hemorrhage. The range of pixel intensities represented by these tissue components was determined by analyzing 250 known image samples of these tissues (Figure 1A). Using these tissue pixel ranges, we identified and mapped pixels in the plaque image to localize and quantify each tissue (Figure 1B) using an image segmentation algorithm (Image Pro-3.1, Media Cybernetics, Silver Spring, MD). PDA was used to predict the amount of plaque blood, lipid, fibromuscular tissue, and calcium from preoperative US images. These correlated closely with histologic quantification performed on 418 slides from the same plaques explanted after CEA.3 Measurements of lipid core size and location with respect to the blood-flow lumen also correlated with corresponding measurements made on histology.8
DIFFERENTIATING SYMPTOMATIC VERSUS ASYMPTOMATIC PLAQUES
PDA was used to quantify tissue (intraplaque hemorrhage, lipid, fibromuscular tissue, and calcium) and architectural features (lipid core size and location) of carotid plaques in 45 arteries prior to undergoing CEA.9 We found that the protocol identified significant differences between symptomatic and asymptomatic plaques (Figure 2). Using PDA, 18 symptomatic plaques demonstrated larger quantities of intraplaque hemorrhage (P<.001) and lipid (P=.002), and larger lipid cores (P=.005) located closer to the flow lumen (P=.01) (Figure 3A, B). The 27 asymptomatic plaques evaluated in the study demonstrated smaller amounts of calcium (P<.001). There was no difference in fibromuscular tissue between the two groups (P=.35). Reproducibility was confirmed, with low inter- and intraobserver variability when different investigators performed PDA on the same plaques.8 Therefore, US image analysis demonstrated that intraplaque hemorrhage, large and superficially located lipid cores, and reduced intraplaque calcification, are potential markers for high risk of stroke in carotid atherosclerotic plaques.
THE FUTURE IS NOW
Increasing awareness regarding the contribution of carotid plaque instability to neurologic complications, the small benefit derived from revascularization of all patients with asymptomatic stenosis, and the potential for plaque-stabilizing pharmacotherapy are all major imperatives driving clinical investigations in this field. MRI has been the only investigated option for accurate morphologic assessment, but cost considerations may limit serial assessments and universal clinical utilization. Improved US imaging can resolve these limitations. Even with an overall brightness assessment of the entire image, hypoechoic plaques were observed to have a small but significantly increased neurologic complication rate during CAS.10 The specific tissue composition and architectural information provided by PDA has the potential to enhance sensitivity and specificity for identifying high-risk lesions for CAS and CEA.11 This is currently being investigated at the authors' institution, as are refinements of the technique to incorporate pixel relationships (textural analysis). Further evaluation in large longitudinal studies will determine its efficacy in stratifying risk according to the method of revascularization, and in following changes in plaques of patients on pharmacologic treatment.
Preoperative PDA offers a unique opportunity to evaluate the specific tissue composition and architecture of carotid plaques. The method correlates well with postoperative histologic analysis. Previous US methodologies provided general composite descriptors with limited correlative histology. The protocol utilizes B-mode imaging, a capability that is readily available in all vascular laboratories. If integrated into routine US software, it could be automated and performed with an additional 10 minutes added to a routine carotid protocol. It provides an inexpensive but rapid real-time assessment of carotid plaque morphology. PDA can identify differences in morphology between symptomatic versus asymptomatic plaques. Histologic studies implicate these features to be precursors for atheroembolization; therefore, these are potential PDA markers for high risk of stroke. Computer-assisted image analysis of B-mode plaque images can revolutionize treatment for carotid artery disease.
Brajesh K. Lal, MD, is Assistant Professor of Vascular Surgery and Physiology at New Jersey Medical School in Newark, New Jersey and Assistant Professor of Biomedical Engineering at Stevens Institute of Technology in Hoboken, New Jersey. He has disclosed that he holds no financial interest in any product or manufacturer mentioned herein. Dr. Lal may be reached at (973) 972-3736; firstname.lastname@example.org.
Robert W. Hobson II, MD, is Professor of Vascular Surgery and Physiology at New Jersey Medical School in Newark, New Jersey. He has disclosed that he holds no financial interest in any product or manufacturer mentioned herein. Dr. Hobson may be reached at email@example.com.
Funding for this research was provided in part by the American Heart Association (BKL), the American College of Surgeons (BKL), and the National Institutes of Health (RWH).