Enhanced Denoising of Ultrasonic Attenuation Images Through Robust Joint Reconstruction

Abstract

The attenuation coefficient slope (ACS) is a parameter used in quantitative ultrasound for tissue characterization. A previous study proposed a multi-frequency framework (WTNV-SLD) for the joint denoising of the spectral ratios by exploiting structural similarities, using a weighted total nuclear variation to improve the quality of the ACS images. This study introduces RobTNV-SLD, a spatially robust estimation method to enhance the denoising of ultrasonic attenuation images, particularly under non-homogeneous conditions such as variable brightness, by incorporating spatial prior and adaptive channel weighting applied with a Lorentzian M-estimator. Metrics were compared to the WTNV-SLD with data from simulated and tissue-mimicking phantoms considering hypoechoic and hyperechoic inclusions. Both techniques reported a comparable estimation bias less than 15% in the simulation and tissue-mimicking phantoms. Nonetheless, in the simulation, RobTNV-SLD achieved a lower root mean square error on the axial profile than WTNV-SLD of 0.194 vs 0.284, reducing the artifacts in boundaries. In the tissue-mimicking phantom, RobTNV-SLD yielded a lower RMS in the axial profile of 0.271 vs 0.409. Thus, providing a superior differentiation of inclusion and background and improved robustness against outliers as artifacts related to non-constant backscatter values and boundary regions.