Combination of ARFI Excitation Powers and Acquisitions at Diastole and Systole for Improving Automatic Segmentation of Vulnerable Carotid Plaque Features

Delineating carotid plaque components that confer rupture risk is vital to stoke prevention. We have previously shown that combining temporal profiles of Acoustic Radiation Force Impulse (ARFI)-induced displacement, cross-correlation coefficient, and signal-to-noise ratio (SNR) as the input feature set to a machine learning classifier enabled the classifier to differentiate intraplaque hemorrhage (IPH), lipid-rich necrotic core (LRNC), collagen (COL), and calcium (CAL) plaque components, with accurate fibrous cap thickness measurement. We hypothesize that machine learning-based classification of carotid plaque structure and composition will be improved by including more information in the input feature set. This study analyzed six carotid plaques imaged in vivo in patients undergoing carotid endarterectomy CEA). Data were acquired with ECG gating to diastole and to systole, with and without ARF excitation. From these data, temporal profiles of displacement, crosscorrelation coefficient (CC), and signal-to-noise ratio (SNR) were generated and used as inputs to a support vector machine classifier. The classifier was trained and tested using spatially matched histology. Over all examined plaques, combining acquisitions at systole and diastole, with and without ARFI push, achieved CNRs that were statistically higher than any one or combination of two inputs. These results suggest that the performance of a machine learning classifier for automatic plaque feature delineation can be improved by optimizing the feature set, with clinical relevance to in vivo human carotid plaque feature delineation herein demonstrated.