Machine Learning Approaches for Quantitative Viscoelastic Response (QVisR) Ultrasound

• Published in: IEEE International Ultrasonics Symposium (Online) pp. 1 - 3
• Main Authors: Richardson, Joseph B., Moore, Christopher J., Anand, Keerthi S., Yokoyama, Keita A., Gallippi, Caterina M.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 07.09.2020
• Subjects: Acoustic Radiation Force (ARF); Acoustics; Elasticity; Elastography; Force; Machine learning; Predictive models; Ultrasonic imaging; Viscoelastic Response (VisR); Viscoelasticity; Viscosity
• ISSN: 1948-5727
• DOI: 10.1109/IUS46767.2020.9251830

We present a quantitative extension of Viscoelastic Response (VisR) ultrasound that estimates shear elastic and viscous moduli from on-axis VisR displacement profiles in silico. Isotropic, homogeneous, linearly viscoelastic materials ranging from 1.57-33.33 kPa shear elasticity and 0.0033-2.34 Pa.s shear viscosity subject to a VisR beamsequence at 26 focal depths were simulated. Multi-target regression machine learning models were used to estimate shear elasticity and shear viscosity given the displacement profile, focal depth, and axial depth. The best performing models achieve a shear elasticity RMSE of 0.29 kPa and a shear viscosity RMSE of 0.13 Pa.s when predictions were made on the test set. These results suggest that machine learning methods can be used to quantitatively estimate viscoelasticity from VisR displacement profiles.