Frequency estimator to improve H-scan tissue characterization

H-scan is a promising quantitative ultrasound method that uses a red/blue color display to illustrate the frequency of backscattered signal, representative of tissue properties. It is based on matched filters of different frequencies, but has a noisy pattern due to the broad spectral distribution of ultrasound. We suggest an improved adaptive frequency estimator reducing the noise in homogeneous regions but maintaining sharpness at boundaries.The proposed frequency estimator utilizes 2D autocorrelation with a matched filter. This estimator provides an a priori estimate of the spatial distribution of frequencies using a matched-filter-based frequency estimation. The heterogeneity of the estimated frequencies is then used to define a 2D weighting function for a second frequency estimation stage. Autocorrelation, borrowing the concept of Loupas’s blood velocity measure, is then used to measure the axial frequency components with a 2D spatial kernel. We use a weighted summation in the autocorrelation based on the spatial frequency distribution within the 2D kernel. The proposed frequency estimator was compared with currently available frequency estimators, including short time Fourier transform, H-scan matched filter, and autocorrelation. We examined the performance of the frequency estimators using Field II simulations and in human subjects with hepatic steatosis.Field II simulation results showed that our adaptive estimator can reduce the noisy texture pattern in H-scan for homogeneous regions while maintaining delineation of boundary interfaces, whereas other estimators are only capable of reduce the noise or maintaining boundary delineation. We applied our adaptive estimator to in vivo human liver, demonstrating improved visualization of H-scan images while reducing the noisy texture pattern and improving differentiation of steatotic liver from gallbladder/skin layer.