Adaptive beamformers for ultrasound imaging of acoustically hard tissues

Abstract

The imaging of bone structures is usually done using X-ray based modalities. However, the application of these modalities can be limited due to unwanted ionizing radiation exposure, scanning cost, and lack of portability. Ultrasound addresses these issues, offering a modality without any known harmful effects. Ultrasound imaging of the bone surface has been investigated in different clinical procedures, e.g., guidance for minimal invasive (MI) procedures in spinal surgery, and for administration of spinal anesthesia. In general, bone imaging using conventional ultrasound techniques is prone to higher level of artifacts in comparison with soft tissue imaging. In the case of the spine, images are filled with acoustical noise, and artifacts that can impede visualization of important features, and also make it hard to detect the bone surfaces. The acoustical noise that appears in images of the bone (vertebra) is partly due to the obstruction of ultrasound beams, and to off-axis signals. The former can degrade the resolution in the lateral direction, resulting in unclear and stretched boundaries of the vertebrae. The latter is due to the sidelobe levels of ultrasound beams, causing unwanted speckles in the shadow region.

The main objective of this thesis is to enhance the visualization of bone surfaces in ultrasound images. This can be beneficial for applications in which extraction of the bone anatomy from B-mode images is of interest. Because the achievable resolution and sidelobe levels are fundamentally limited in the standard delay-andsum beamforming technique, we investigate the potential of adaptive beamformers to alleviate some of the acoustical noise observed in the related images, and to improve the bone edges in ultrasound images.

In Paper A, we address artifacts (shadowing effects) resulting from partial obstruction of the imaging aperture in bone imaging scenarios. We investigate the potential of the minimum variance (MV) beamforming method to alleviate these artifacts. We show that the robustness of the MV beamformer degrades when the imaging aperture is highly obstructed by the bone structure due to the weak estimation of the covariance matrix. We suggest that the covariance matrix has to be estimated based only on the data from the un-shadowed elements. Thus, we adaptively determine the shadowed elements and discard their corresponding data from the covariance matrix to improve the MV beamformer performance.

In Paper B, we follow two main goals: to investigate the robustness of the minimum variance based beamformers in bone imaging scenarios, and to study an eigenspace minimum variance beamformer (ESMV) to improve the edges of the acoustically hard tissues in the ultrasound images. In this paper, we use forward/backward averaging to enhance the covariance matrix estimation in imaging scenarios in which shadowing may occur. The enhanced covariance matrix is used to estimate ESMV weights. We show that the performance of the ESMV beamformer depends on the estimation of the signal subspace rank. The lower ranks of the signal subspace can enhance the edges and reduce noise in ultrasound images; however, the speckle pattern can become distorted.

In Paper C, we investigate the potential of a framework for extracting the bone surface from B-mode images. In this framework, we use the ESMV beamformer technique together with a feature detection method as a tool for extracting the bone surfaces. In this paper, we show that an ESMV beamformer with a rank-1 signal subspace can preserve the bone anatomy and enhance the edges reasonably well, despite some distortion of the speckle pattern. Also, this makes the beamformer independent of the signal subspace estimation, which is one of the limitations of the eigenspace beamformers. The beamformed images are post-processed using a feature detection technique, and here we use the phase symmetry (PS). This method utilizes 2D Log- Gabor filters and has been shown to be effective as a ridge detector in bone localization in US images. We examine the proposed framework for imaging the spinal anatomy.