Homodyned K-distribution parametric maps combined with elastograms for carotid artery plaque assessment

摘要

The goal of this work was to perform tissue characterization of atherosclerotic carotid plaques to assess their vulnerability to rupture. It was hypothesized that the echo envelope of radiofrequency signals of carotid arteries at 7.2 MHz is distributed locally according to homodyned K-distributions (HKD). Based on the statistical parameters of this distribution, two HKD parametric maps could be constructed, a first one related to the mean intensity and the second one related to the diffuse intensity SNR2, which is a novel feature in this context. The objective was to test if quantities extracted from such parametric maps, together with elastograms, could distinguish between non-vulnerable carotid artery plaques, and neovascularized or vulnerable ones, as assessed with magnetic resonance imaging. This novel combination of mean intensity, diffuse intensity SNR2, and elastogram, was quantitatively tested using a neural network with two hidden layers of 21 and 7 neurons, trained by minimizing weighted cross entropy, on plaques of 31 patients (12 non-vulnerable plaques, 19 neovascularized or vulnerable ones). Sensitivity and specificity were estimated with a bootstrap method. Classification based solely on elastograms yielded a specificity of 66.6% for a sensitivity of 82.7%. For the same sensitivity, the specificity was raised to 79.3% with the proposed combined method.