Attenuation coefficient imaging using regularization by denoising

The attenuation coefficient (AC) is parameter used in quantitative ultrasound that allows the characterization of different tissue types. The regularized spectral log difference (RSLD) method is a technique that uses a regularization step based on total variation in order to extend the trade-off between the estimation variance and spatial resolution. However, the RSLD method consider only piecewise homogeneous media, an assumption that may not hold true in clinical applications. Therefore, it is necessary to study alternative regularization methods for stabilizing attenuation imaging methods. This work introduces a new regularization method based on regularization by denoising (RED) to obtain improved estimates of ACs. This algorithm is validated using computer simulations, physical phantoms and an in vivo thyroid sample. Whereas the performance of the RSLD and RED-based methods were comparable with simulated media, the RED-based method outperformed RSLD with physical phantoms reducing the coefficient of variation by nearly a factor of 3 while maintaining the same accuracy. Improvements were also observed with the in vivo dataset, where the mean value estimated with RSLD was highly sensitive to the selection of the region of analysis, experiencing nearly a twofold variation. In contrast, the RED-based method provided mean estimated ACs with less than a 20% variation. These results suggest that the proposed method may exhibit a greater robustness when estimating attenuation coefficients than their total variation based counterparts.