David P. Duncan

Classifying ultrasound image regions by using characteristics of the ARFI induced tissue displacement temporal profile

Classifying an ultrasound image into regions with diagnostic information can directly assist clinicians and can also be used as a preprocessing stage in parametric image construction. In this work we propose to extract information from an Acoustic Radiation Force Impulse (ARFI) induced displacement temporal profile and use this information as input to a linear classifier to assign each image sample to one category, either fluid, high stiffness, low stiffness or undetermined. Three parameters were derived from the displacement profile, (1) signal to noise ratio (SNRD), (2) maximum displacement (Dmax), and (3) time-to-peak displacement (TTP). The proposed method was tested on phantoms which contain a fluid filled cyst and solid inclusions with various stiffness values, two diagnostically confirmed human breast simple cysts, and five Microwave ablated thermal lesions in ex-vivo bovine livers. Using 35.6kPa as separation between low and high stiffness, for phantom and breast simple cysts we obtained a mean classification accuracy of 93%; for thermal lesions, we obtained a mean classification accuracy of 81%.

Tracking ARFI induced shear wave propagation by using maximum likelihood estimation

Shear wave propagation speed in tissue provides valuable diagnostic information to clinicians and has been investigated in many clinical studies. While shear velocity measurement based on acoustic radiation force impulse (ARFI) imaging produces reliable results in homogeneous media, it is still a challenge to accurately detect the shear wave propagation velocity in heterogeneous tissue. The reflection of the shear wave in heterogeneous media can severely alter the displacement temporal profile shape from location to location and makes detection methods perform poorly when estimating shear wave propagation time. In this work a parametric model was developed to minimize the affect of altered displacement profile on the estimate. The new method uses shear velocities between the excitation and detection spaces as model parameters and uses the normalized displacement temporal profiles at detection locations as probability distribution functions. A joint density (likelihood) function is then constructed based on the shear velocities and the displacement profiles. Searching for the maximum value of the likelihood function by using an iterative method produces estimates of the shear velocity which are more robust to shear wave reflections.