Towards ultrasound full-waveform inversion in medical imaging

Ultrasound imaging is a front-line clinical modality with a wide range of applications. However, there are limitations to conventional methods for some medical imaging problems, including the imaging of the intact brain. The goal of this thesis is to explore and build on recent technological advances in ultrasonics and related areas such as geophysics, including the ultrasound data parallel acquisition hardware, advanced computational techniques for field modelling and for inverse problem solving. With the significant increase in the computational power now available, a particular focus will be put on exploring the potential of full-waveform inversion (FWI), a high-resolution image reconstruction technique which has shown significant success in seismic exploration, for medical imaging applications. In this thesis a range of technologies and systems have been developed in order to improve ultrasound imaging by taking advantage of these recent advances. In the first part of this thesis the application of dual frequency ultrasound for contrast enhanced imaging of neurovasculature in the mouse brain is investigated. Here we demonstrated a significant improvement in the contrast-to-tissue ratio that could be achieved by using a multi-probe, dual frequency imaging system when compared to a conventional approach using a single high frequency probe. However, without a sufficiently accurate calibration method to determine the positioning of these probes the image resolution was found to be significantly reduced. To mitigate the impact of these positioning errors, a second study was carried out to develop a sophisticated dual probe ultrasound tomography acquisition system with a robust methodology for the calibration of transducer positions. This led to a greater focus on the development of ultrasound tomography applications in medical imaging using FWI. A 2.5D brain phantom was designed that consisted of a soft tissue brain model surrounded by a hard skull mimicking material to simulate a transcranial imaging problem. This was used to demonstrate for the first time, as far as we are aware, the experimental feasibility of imaging the brain through skull using FWI. Furthermore, to address the lack of broadband sensors available for medical FWI reconstruction applications, a deep learning neural network was proposed for the bandwidth extension of observed narrowband data. A demonstration of this proposed technique was then carried out by improving the FWI image reconstruction of experimentally acquired breast phantom imaging data. Finally, the FWI imaging method was expanded for3D neuroimaging applications and an in silico feasibility of reconstructing the mouse brain with commercial transducers is demonstrated.Open Acces