Abstract:
Estimation of ultrasonic attenuation coefficient (AC) is essential for quantifying and characterizing the features of tissue microstructure. In the conventional AC estima- tion methods, a well-specified reference phantom is commonly used for minimizing the diffraction and transmit pulse related effects on the ultrasound radio-frequency (RF) echo signal. In this thesis, a novel AC estimation technique is proposed avoiding the need of using any reference data, where the undesired system effects on the RF data are minimized through point spread function (PSF) separation and band-pass filtering of the envelope signal of the tissue reflectivity function (TRF). An improved and compu- tationally efficient non-parametric cepstrum-based technique is used for separating the TRF and PSF from the measured RF signal. The Hilbert transform based temporal envelope is introduced to smooth out the unwanted effects of discontinuity and noise in the TRF and PSF signals. Finally, a band-pass filter based log power approximation technique is applied to estimate the center frequency component of the attenuating TRF envelope power spectra with reduced diffraction effect. Assuming continuity of AC within a small uniform region, an exponentially weighted-average of logarithmic signal power of the neighboring blocks at the center frequency is measured for different depths, with a view to fit a regression line for obtaining an average AC value from its slope. Comparative results of the proposed reference-free AC estimation method with other conventional reference-based methods are presented for tissue-mimicking (TM) phantoms, in vivo breast and liver data. For the TM phantoms, the AC estimates using the proposed algorithm are within 10% deviation of the actual values. The obtained results for the normal female breasts, normal human livers, and fatty livers are 0.44 ± 0.23 dB/cm-MHz, 0.55 ± 0.21 dB/cm-MHz, and 0.61 ± 0.20 dB/cm-MHz, respectively, which are consistent with the literature-reported values of AC. Different from reference-based methods, the proposed technique is free from the bias that may result from the dissimilarity between acoustic characteristics of the reference and sample.
