Super-Resolution Ultrasound Imaging of Microvascular Networks
Abstract
Under development in a pre-clinical setting, super-resolution ultrasound (SR-US) imaging brings up to a ten-fold improvement in ultrasound (US) resolution. It can image microvascular changes in various disease processes including cancer, diabetes, and cardiovascular disease. Barriers to the clinical use of SR-US include the heavy computational burden of both detecting and localizing the microbubble (MB) contrast agent, as well as lengthy US imaging times longer than a breath hold. The long computation time of minutes or hours requires the use of offline processing and precludes the use of SR-US as a real-time imaging modality which is viewed as a major advantage of US imaging. We hypothesize that the use nonlinear multi-pulse sequences will improve the numbers of MBs detected per frame and reduce image acquisition time for SR-US towards a comfortable breath-hold length for the patient. We further hypothesize that the use of deep learning algorithms can begin to enable a real-time SR-US imaging modality by the reduction of computation time by several orders of magnitude. In this dissertation work we demonstrate: (1) improved MB detection in SR-US with the use of nonlinear US pulse sequences to enhance the contrast of MBs with tissue, (2), improve performance of super-resolution processing with deep learning for MB detection and precise localization, and (3) equivalent performance of deep learning methods to current methods in the assessment of tumor treatment by monitoring any microvessel changes.