To determine the feasibility of texture analysis for the classification of American Heart Association (AHA) type IV-V and type VI carotid atherosclerotic lesions at multi-contrast black-blood MR images.

Patients and Study Design

A cross-sectional study was performed on 24 patients with carotid stenosis >50% detected by ultrasound. All subjects underwent a standard multi-contrast MR vessel wall imaging protocol. MRI-modified AHA criteria¹ was used to evaluate carotid atherosclerotic lesion type and each cross-sectional location was evaluated. The main criterion for inclusion in the study is the presence of type IV-V and type VI carotid atherosclerotic lesions (type IV–V represents plaque with a lipid or necrotic core surrounded by fibrous tissue with possible calcification; type VI represents complex plaque with a possible surface defect, hemorrhage, or thrombus). Image quality was rated per axial location on a four-point scale (1, poor; 2, marginal; 3, good; 4, excellent) depending on the overall signal-to-noise ratio and the clarity of the vessel wall boundaries. Slices with an image quality < 2 were excluded from review.

Carotid MRI Protocol

All patients were imaged using a 3.0T MRI scanner (Philips Achieva, Best, Netherlands) with an 8-channel phased-array carotid coil. A standardized imaging protocol was performed to obtain multi-contrast cross-sectional MR images, including time-of-flight (TOF), T1W, T2W, MP-RAGE and contrast-enhanced (CE)-T1W imaging of the bilateral carotid arteries centered on the bifurcation described before². Gadodiamide (GE Healthcare) was injected intravenously at a dose of 0.1 mmol/kg and a rate of 1 mL/s 5 minutes prior to CE-T1W image acquisition. All MRI axial scans were acquired with a section thickness of 2 mm, a FOV of 14 x 14 cm, a matrix size of 256x256, and an in-plane resolution of 0.54-0.55 mm.

Texture Analysis and Lesion Classification

We used the popular MaZda 4.6 texture analysis software³ to generate reproducible regions of interest (ROIs) for all type IV-V and type VI carotid lesions on T1W, T2W and CE-T1W images. For each lesion, a single free-hand ROI of plaque was manually depicted by an experienced radiologists blind to the clinical information and carotid lesion AHA classification. Then, texture features derived from the gray-level histogram, co-occurrence and run-length matrix, gradient, autoregressive model, and wavelet transform were calculated. To identify the most valuable texture features for distinguishing between type IV-V and type VI lesions, top 3 subsets of each 10 texture features were extracted, independently for T1W, T2W and CE-T1W sequences. Fisher, probability of classification error and average correlation (POE + ACC), and mutual information (MI) coefficients were used to extract subsets of optimized texture features. Linear discriminant analysis (LDA) was used for lesion classification.

Four patients showed considerable artifacts on carotid MR images. Of the remaining 20 subjects, 16 were male (80 %) and the mean age was 62.1 ± 10.2 years. In total, 23 arteries had 26 location with type VI lesions, 31 location with type IV-V lesions. Of the arteries with AHA- type VI, 88.5% had hemorrhage and 11.5% had a ruptured fibrous cap. On pre- and post-contrast T1-weighted images, top 3 texture features that were selected based on the Fisher and MI coefficients were predominantly derived from the co-occurrence matrix, whereas feature subsets created using the POE + ACC method included features from all categories except the run-length matrix (Table 1). LDA produced misclassification rates of 1.75–8.77% on T1W, 24.56–36.84% on T2W images, and 26.32–35.09% on CE-T1W images (Table 2). The discrimination distribution of the three statistical analysis approaches is illustrated in Figure 1, which shows that POE + ACC on T1W has the largest discrimination power to distinguish the two texture categories.MRI texture analysis of carotid atherosclerotic lesions, to our knowledge, has not been previously investigated. Our results suggest that the standard spatial resolution of carotid plaque MR images at 3.0T is sufficient for the calculation of texture features that enable differentiation of type VI lesion (vulnerable plaque) from type IV-V lesion (mostly stable plaque) in the majority of cases (>91% for LDA classification on T1W). However, these two type lesions cannot be reliably distinguished from each other on T2W or with contrast media application, which may result from the fact they share fundamental MRI characteristics^4,5: a hypo–intense on black-blood T2W and iso-intense on the black blood CE-T1W. Texture analysis may be of benefit when used in conjunction with other carotid plaque interpretation like morphological and compositional characteristics for identification of a high-risk vulnerable plaque.Our study results suggest that texture-based pattern classification of advanced carotid atherosclerotic lesions is feasible on MR images, even without the administration of contrast media. This preliminary evaluation indicates that carotid plaque texture analysis is a potentially useful adjunct tool for comprehensive and quantitative evaluation of atherosclerotic plaque vulnerability.