Study On Eigenspace-based Minimum Variance Beamforming In Medical Ultrasound Imaging

Adaptive beamformer has been widely studied in recent years. Some methods are proposed, which brings new opportunities in improving the quality of medical ultrasound imaging. Among these methods, the eigenspace-based minimum variance (ESBMV) beamformer has great potential in medical imaging for its good imaging resolution and contrast. The ESBMV beamformer constructs two orthogonal subspaces, the signal subspace and the noise subspace. By projecting the echo signal onto the signal subspace, the ESBMV beamformer can greatly suppress the power of the noise and acquire the accurate estimation of the signal power. However, the ESBMV beamformer has three deficiencies, which limits its practical implementation. In this dissertation, three improved methods are proposed to cope with the three corresponding deficiencies.1. ESBMV beamformer combined with Wiener postfilter (ESB-Wiener) is proposed in order to solve the deficiency that the imaging resolution of the ESBMV beamformer is almost the same with the minimum variance (MV) beamformer. In this method, the output of the ESBMV beamformer is optimized with the Wiener postfilter, which makes the output power of the beamformer closer to the actual signal power. For those imaging points with large noise, the ESB-Wiener beamformer can further reduce the noise power. The performance of the new method is demonstrated by resolving point scatterers and cyst phantom. The results show that the new beamformer can achieve higher imaging resolution and contrast without adding too much caculation. In addition, the effect of the sound speed error is investigated by artificially overestimating the speed used in calculating the propagation delay. The results show that the new beamformer provides better robustness against the sound speed errors.2. Beam-domain ESBMV (B-ESBMV) beamformer is proposed in order to solve the deficiency that the computational complexity of the ESBMV beamformer is much higher than that of the MV beamformer. In this method, the array-domain signal is transformed to the beam-domain signal which is then used to form the final output instead of the array-domain signal. With the appropriate transformation, the power of the beam-domain signal can be centralized in certain beams. By keeping only the beams with most power to form the output of the beamformer, the dimension of the covariance matrix can be reduced. Thus the computational complexity of the beamformer can be reduced. The performance of the new method is demonstrated by resolving point scatterers and cyst phantom. The results show that the B-ESBMV beamformer can achieve almost the same performance as the ESBMV beamformer in much shorter time.3. Target-finding ESBMV (T-ESBMV) beamformer is proposed in order to solve the deficiency that the ESBMV beamformer brings distorted results of "dark area" around the strong scatterers. The reason for this distortion is studied and a modified method of constructing the signal subspace is proposed. By adding the eigenvectors which are closely related to the steering vector into the original signal subspace, the T-ESBMV beamformer can acquire a more accurate signal subspace. The performance of the new method is demonstrated by resolving point scatterers and cyst phantom. The results show that the T-ESBMV beamformer can effectively eliminate the distortion around the strong scatterers.