Spectral Analysis of Nonlinearly Generated Second Harmonic Backscatter for Characterization of Human Carotid Plaque

Plaque composition is generally unavailable for patients with carotid stenosis. The objective of this research is to develop a non-invasive approach for determining the composition of carotid plaque in order to improve risk stratification. Human carotid plaque is collected following carotid endarterectomy from 46 subjects. Serial histology is performed and the slides are matched to grayscale images created from RF data collected in vivo prior to surgery. Regions of homogenous tissue (F-fibrous, HNC - hemorrhagic and/or necrotic core, Ca - calcium) are determined based on histology. The corresponding RF data is processed by computing an average power spectrum over a 1.2 mm by 1.2 mm region (15 lines by 64 points) and normalized using a reference phantom with a 0.5 dB /cm-MHz attenuating phantom. An additional attenuation compensation is applied based on backscatter from healthy volunteer adventitia as a function of depth and frequency. Random forests were created using eight spectral parameters for both fundamental and harmonic bands. The out of bag error is 0.26, 0.25, and 0.23 for forests based on fundamental data, harmonic data, and combined data sets.