Multiparametric attenuation and backscatter images for characterization of carotid plaque

The goal of this study was to develop methods for quantitative ultrasound imagery suitable for noninvasive assessment of carotid plaque composition prior to the selection of the technique for revascularization. Using two broadband transducers (5-12 MHz and 12-28 MHz), backscattered radio frequency signals were acquired from entire lengths of 15 carotid endarterectomy specimens. Spectral analysis methods with correction for system response and beam diffraction were applied to radio frequency signals from local volumes of plaque having a 2 mm slice thickness, 1 mm width and axial depth of 480 microm and 240 microm at 10 MHz and 20 MHz, respectively. From these spectra, local values of four ultrasound parameters (integrated backscatter, frequency dependence of backscatter, integrated attenuation and slope of attenuation) were estimated and used to construct quantitative images. To combine information from these different parameter images, a two-step approach was followed. First, in 59 independent quantitative images of highly stenotic plaque, the average parameter values in a central five-by-three pixel region were correlated with plaque composition as assessed by histology to investigate the relationship between parameter values, frequency bandwidth and plaque composition. Discriminant analysis of parameter values vs. plaque composition was made to find a set of predictive equations to classify sets of measurements. Correct classification was obtained for 100% of calcified, 75% of intraplaque hemorrhage and 71% of lipidic plaques of the input data set. Second, each set of pixels from different parameter images was classified using the predictive equations, and a single, local tissue composition image was constructed. Examples of tissue composition images are presented in comparison with corresponding histologic sections. Both agreement and disagreement between image pairs are discussed.