Prediction-error filters for signal and noise separation within ultrasound recordings

Ultrasonic image quality can be compromised by several mechanisms such as reverberation, off-axis scattering, and sound-speed inhomogeneities. These mechanisms can cause a loss of coherence of signals across the aperture and clutter in a B-mode image potentially leading to inadequate visualization and misdiagnosis. We develop a processing technique based on prediction-error filters (PEFs) for isolating tissue signal and several types of noise in the individual channel data of ultrasound scans. We first estimate a set of PEFs that allows for prediction of the 2-D spectra of signal and noise in the channel data domain and then use these filters to separate the signal and noise components within the recorded echoes. We use the Van Cittert-Zernike theorem as a priori knowledge of the signal spectrum to develop the signal-PEF, while the noise-PEFs are estimated directly from the channel data. We apply our PEFs to channel data from fullwave simulations of abdominal scans and to data acquired on a lesion phantom through a layer of bovine tissue. The proposed technique successfully separates signal from noise and improves lag-one coherence of the speckle signal from 0.47 to 0.97.Ultrasonic image quality can be compromised by several mechanisms such as reverberation, off-axis scattering, and sound-speed inhomogeneities. These mechanisms can cause a loss of coherence of signals across the aperture and clutter in a B-mode image potentially leading to inadequate visualization and misdiagnosis. We develop a processing technique based on prediction-error filters (PEFs) for isolating tissue signal and several types of noise in the individual channel data of ultrasound scans. We first estimate a set of PEFs that allows for prediction of the 2-D spectra of signal and noise in the channel data domain and then use these filters to separate the signal and noise components within the recorded echoes. We use the Van Cittert-Zernike theorem as a priori knowledge of the signal spectrum to develop the signal-PEF, while the noise-PEFs are estimated directly from the channel data. We apply our PEFs to channel data from fullwave simulations of abdominal scans and to data acquired on a lesion phantom...