Assessment of human atherosclerotic carotid plaque components with multisequence MR imaging: initial experience.

Abstract

To prospectively determine, by using a stepwise logistic regression model, the optimal magnetic resonance (MR) weighting (ie, pulse sequence) combinations for plaque assessment and corresponding cutoff values of relative signal intensities (rSIs). Institutional review board approval and patient consent were obtained. Eleven patients (seven men, four women; mean age +/- standard deviation, 68 years +/- 4) with symptomatic carotid disease and stenosis of more than 70% were investigated at MR imaging before carotid endarterectomy. The MR images were matched with histologic features of the endarterectomy specimens (reference standard). The rSIs (compared with that of muscle tissue) from regions of interest were assessed qualitatively and semiquantitatively. For all major components (calcification, lipid core, intraplaque hemorrhage, and fibrous tissue), optimal cutoff points for the rSIs were determined for five MR weightings by means of receiver operating characteristic curves. The best predicting combinations of these five dichotomized MR weightings were selected by means of stepwise logistic regression analysis. The potential sensitivity and specificity of MR imaging for vulnerable plaque with hemorrhage and/or lipid core were determined. The same optimal MR weighting combinations for identifying the four plaque components were found with qualitative and semiquantitative analysis. Sensitivity and specificity for vulnerable plaque were 93% (95% confidence interval: 77%, 99%) and 96% (95% confidence interval: 86%, 100%), respectively, for the qualitative analysis and 76% (95% confidence interval: 56%, 90%) and 100% (95% confidence interval: 93%, 100%) for the semiquantitative analysis. This study demonstrates the potential of a systematic approach of atherosclerotic plaque assessment with multisequence MR imaging by using the information provided from five different MR weightings in a stepwise logistic regression model.