Attenuation Compensation Comparison for Human Carotid Plaque Characterization Using Spectral Analysis of Backscattered Ultrasound

Abstract

Carotid atherosclerotic plaque composition is a valuable predictor of stroke risk. Ultrasound spectral analysis has been successfully implemented clinically for determining plaque composition in coronary arteries via intravascular ultrasound. Noninvasive implementation for carotid plaque requires compensation for the attenuating effects of overlying tissue. This study examines the effects of four attenuation compensation techniques on the accuracy of a carotid plaque classification system using spectral analysis and random forest machine learning classification. Radiofrequency (RF) data was acquired from 20 subjects prior to carotid endarterectomy (CEA). 41 fibrous (F), 60 hemorrhagic and/or necrotic core (H/NC), and 54 calcified (Ca) regions of interest (ROI) were selected from the RF data corresponding to homogenous zones within the histology of the excised plaque tissue. Additionally, 219 ROI’s were obtained from the adventitia (Adv) of six normal subjects. Power spectra for the ROI’s were computed and normalized to a uniform phantom. Four attenuation compensation methods were applied to the spectra: (1) 0.5 dB/cm-MHz; (2) optimum power spectral shift estimator (OPSSE); (3) 1-step and (4) 2-step normalized backscatter from adventitia. A linear fit of the resulting estimated backscatter transfer functions (eBTF) was performed over the fundamental bandwidth of 2.5 – 6.9 MHz. Eight spectral parameters were used to build the random forest classification models. While there were no statistically significant differences in the accuracy of the classification models based off each attenuation compensation approach, our work has shown that additional attenuation compensation may provide a benefit for characterizing carotid plaque.