Spatial Prediction Filtering of Acoustic Clutter and Random Noise in Medical Ultrasound Imaging

One of the major challenges in array-based medical ultrasound imaging is the image quality degradation caused by sidelobes and off-axis clutter, which is an inherent limitation of the conventional delay-and-sum (DAS) beamforming operating on a finite aperture. Ultrasound image quality is further degraded in imaging applications involving strong tissue attenuation and/or low transmit power. In order to effectively suppress acoustic clutter from off-axis targets and random noise in a robust manner, we introduce in this paper a new adaptive filtering technique called frequency-space (F-X) prediction filtering or FXPF, which was first developed in seismic imaging for random noise attenuation. Seismologists developed FXPF based on the fact that linear and quasilinear events or wavefronts in the time-space (T-X) domain are manifested as a superposition of harmonics in the frequency-space (F-X) domain, which can be predicted using an auto-regressive (AR) model. We describe the FXPF technique as a spectral estimation or a direction-of-arrival problem, and explain why adaptation of this technique into medical ultrasound imaging is beneficial. We apply our new technique to simulated and tissue-mimicking phantom data. Our results demonstrate that FXPF achieves CNR improvements of 26% in simulated noise-free anechoic cyst, 109% in simulated anechoic cyst contaminated with random noise of 15 dB SNR, and 93% for experimental anechoic cyst from a custom-made tissue-mimicking phantom. Our findings suggest that FXPF is an effective technique to enhance ultrasound image contrast and has potential to improve the visualization of clinically important anatomical structures and diagnosis of diseased conditions.