Characterization of Heterogeneous Perfusion in Contrast-Enhanced Ultrasound

Ultrasound contrast agent (UCA) imaging continues to offer the promise of functional imaging with a number of existing clinical applications such as myocardial perfusion. However, imaging UCA in heterogeneous perfusion regions continue to be a challenge. We present an algorithm combining nonlinear imaging, motion compensation and machine learning to improve the specificity of perfusion imaging to UCA activity in vivo. The algorithm also employs singular value decomposition to identify and suppress signal components with high sensitivity to tissue motion, specular reflections and noise. The algorithm used motion compensation, differential filtering, and mode selection to penalize the effects of tissue motion, specular reflection, and noise while rewarding sporadic contrast activity in order to increase sensitivity and specificity to perfusion in the tissue. These modes were fed into a neural network for quantitative classification of the perfusion in the tissue. The results from in vivo imaging of a heterogeneous tumor model exhibit high degree of separation in computed perfusion index values with and without UCA.