Advanced beamforming for high frame rate ultrasound vascular imaging

Abstract

Ultrasound vascular imaging has seen two major advances which are unfolding unprecedented possibilities for research: the first being the use of microbubbles as contrast agents to highlight the vasculature, while the second one is the use of unfocused transmission and digital image reconstruction techniques yielding a remarkable frame-rate increase. Their consolidation is pushing scientists to design acquisition procedures and reconstruction models to surpass conventional approaches developed for focused ultrasound, taking full advantage of the new available data. In this thesis, signal processing, in particular spatial and temporal second order statistics, has been used to try pushing the boundaries in this direction.

In the first contributing chapter, some limitations of classical coherent compounding and multipulse imaging methods for contrast ultrasound are highlighted, with the use of both simulations and in vivo experiments, in particular when fast flow is imaged: as an example, the case of cardiac imaging with diverging waves is reported. The results show how motion is responsible for large intensity drops, up to tens of dB, which can have potential detrimental effects in quantifications studies. The results also show the dependency of this effect on a variety of acquisition settings and we also highlight some artefacts arising from those scenarios. Finally, reduction of such artefacts by correcting motion is demonstrated.

In the second contributing chapter, a new method is developed to reduce noise in High-Frame Rate vascular images. Standard vascular imaging using e.g Power Doppler are characterized by a low Signal to Noise Ratio, especially for unfocused acquisitions. Using cross correlation between two different estimates of the same target vascular signal, we devised a method called Acoustic Sub-Aperture Processing (ASAP) which reduces the noise floor given by random uncorrelated sources, such as electronics, while maintaining the same power estimate. The two estimates can be chosen both as different spatial location within the same point spread function, or as the results of two orthogonal projections of the same pixel value. In this way, a noise reduction proportional to the square root of the number of averaged frames is achieved. The effectiveness of this technique has been tested in several in-vivo settings, both for contrast and non contrast acquisitions.

In the third contributing chapter, methods are developed to reduce imaging artefacts such as side lobes and grating lobes. While effective, ASAP is susceptible to artefacts due principally to the interference generated by strong scatterers like big vessels. Under certain conditions, the presence of artefacts can even reduce the ability to image small vessels compared to simpler approaches. Assuming lack of correlation of the different interference signals, we have devised an adaptive method to select the two sub-apertures in the noise subspace. The results show that such method is effective at mitigating the contribution of interferences to the final image and therefore the strength of the artefacts, improving the visualization of smaller vessels, albeit creating a tradeoff between computational time and quality of the result. Because of the computational load required by the proposed method, we also propose an alternative fast approach based on the spatial Fourier transform.

As ASAP is not the only technique based on cross correlation, the last contributing chapter proposes a tensorial framework capable of the generalizing several coherence methods currently available, by taking into account the different possible domains of correlations, like space, channels, depth and frames. This is done by extending the variance matrix to a variance tensor. Besides deepening the understanding of the interconnection of the various techniques, the multilinear approach allows to combine the various methods to achieve superior performances while keeping the computational time manageable compared to adaptive methods. The algorithms have been preliminary tested in challenging scenarios like contrast enhanced cardiac acquisitions and in vitro. The results suggest improvements in terms of interference and artefacts reductions compared to matrix based processing. A promising future direction is suggested, namely the use of tensorial decomposition techniques for source separation to improve the reconstruction of vascular images