Segmentation of carotid plaque using multicontrast 3D gradient echo MRI

Abstract

Purpose: To evaluate the performance of automatic segmentation of atherosclerotic plaque components using solely multicontrast 3D gradient echo (GRE) magnetic resonance imaging (MRI).

Materials and methods: A total of 15 patients with a history of recent transient ischemic attacks or stroke underwent carotid vessel wall imaging bilaterally with a combination of 2D turbo spin echo (TSE) sequences and 3D GRE sequences. The TSE sequences included T1-weighted, T2-weighted, and contrast-enhanced T1-weighted scans. The 3D GRE sequences included time-of-flight (TOF), magnetization-prepared rapid gradient echo (MP-RAGE), and motion-sensitized driven equilibrium prepared rapid gradient echo (MERGE) scans. From these images, the previously developed morphology-enhanced probabilistic plaque segmentation (MEPPS) algorithm was retrained based solely on the 3D GRE sequences to segment necrotic core (NC), calcification (CA), and loose matrix (LM). Segmentation performance was assessed using a leave-one-out cross-validation approach via comparing the new 3D-MEPPS algorithm to the original MEPPS algorithm that was based on the traditional multicontrast protocol including 2D TSE and TOF sequences.

Results: Twenty arteries of 15 subjects were found to exhibit significant plaques within the coverage of all imaging sequences. For these arteries, between new and original MEPPS algorithms, the areas per slice exhibited correlation coefficients of 0.86 for NC, 0.99 for CA, and 0.80 for LM; no significant area bias was observed.

Conclusion: The combination of 3D imaging sequences (TOF, MP-RAGE, and MERGE) can provide sufficient contrast to distinguish NC, CA, and LM. Automatic segmentation using 3D sequences and traditional multicontrast protocol produced highly similar results.

Figure 1

A comparison of segmentation results using MEPPS and 3D-MEPPS: (a)–(f) Lumen and outerwall displayed on (a) T1, (b) T2, (c) CET1, (d) TOF, (e) MP-RAGE and (f) MERGE; (g) Segmentation results based on MEPPS; (h)Segmentation results based on 3D-MEPPS; Fibrous tissue is white, loose matrix is purple, calcification is blue and necrotic core is yellow.

Figure 2

Visualization of the segmentation results using 3D-MEPPS in (a) axial and (b) longitudinal cross section. Fibrous tissue is white, loose matrix is purple, calcification is blue and necrotic core is yellow. Because the ECA was not analyzed, a discontinuity in the vessel wall (arrow) is present in the longitudinal cross section.

Figure 3

Correlation plots of the segmentation area between MEPPS and 3D-MEPPS;

Figure 4

Joint Probability Distributions of intensities on the three contrast weightings used by 3D-MEPPS. Axes units are relative to the median intensities. Dotted lines correspond to one, two and three times the median intensities. The median intensity generally corresponds to the intensity of fibrous tissue/muscle and is referred to as isointense in manual review.