Multi-parametric Ultrasound Tissue Characterization (MUTC) as a surrogate to Magnetic Resonance Imaging (MRI) for Non-Alcoholic Fatty Liver Disease (NAFLD) characterization

Ultrasound tissue characterization (UTC) provides ultrasound elastography and quantitative metrics extracted from RF data for evaluating tissue mechanics and microstructure. Our goal is to learn multi-parametric models on the same patient population for regression of Magnetic Resonance Imaging Proton-Density Fat-Fraction (MRI-PDFF). The results show that non-linear multi-parameteric UTC models (e.g., multi-layer perceptron, random forests, Gaussian processes) are superior to multi-linear regression approaches and outperform qualitative radiologist assessment. Integration of radiologist feedback with UTC further improves PDFF regression.