Performance of various spectral estimation methods on acoustic backscatter coefficient estimation under data size limitations

The accuracy and precision of in vivo estimates of the backscatter coefficient (BSC) obtained by the Reference Phantom method (RPM) can be affected by limitations on the length of radiofrequency (RF) echo signal segments from which power spectra are obtained. This work compares the performance of four power spectral estimation methods (Short-time Fourier transform with Hann and Blackman-Harris gating functions, Welch periodogram, and Thomson's multi-taper method) when estimating the non-smooth BSC of a tissue-mimicking phantom when the spectral window length is severely reduced. A total of 9 RF echo signal frames were acquired from a tissue-mimicking spherical inclusion and its surrounding background using a clinical ultrasound system. Power spectra were computed with the four methods mentioned above at various locations within each frame by applying spectral analysis windows to RF signal segments ranging from 17 down to 2.4 pulse-echo correlation lengths (PECLs). The RPM was used for estimating the inclusion's BSC with the background as reference. Results show that for windows longer than ~7 PECLs, subwindow-averaging spectral estimation methods, such as the Welch and the multi-taper methods, reduce the variability of the BSC estimates. The use of these methods with shorter windows leads to increased bias, particularly if the BSC is a rapidly varying function of frequency. Therefore, the expected BSC frequency dependence should also be considered when choosing a convenient spectral estimation method.