Ultrafast Medical Ultrasound Imaging Using Diverging Waves: Optimization Strategies for Enhanced Image Quality

Ultrafast imaging has rapidly advanced the field of medical ultrasound by enabling data acquisition and image reconstruction at extremely high frame rates, approximately 1 to 2 orders of magnitude higher than conventional methods. Several advanced applications, such as shear-wave elastography, high-sensitivity Doppler imaging, contrast-enhanced imaging, and functional ultrasound have been made effective by the availability of ultrafast acquisition methods. Ultrafast imaging is typically achieved using either plane waves (PWs) or diverging waves (DWs). Both methods employ all transducer elements for transmission, PWs using varying steering angles, and DWs by spherical waves emitted from virtual sources(VS) placed behind the probe. The use of DWs in ultrafast imaging has two-fold advantages: high frame rate and wide field of view; however, despite these advancements, ultrafast imaging faces fundamental trade-offs—most notably, a compromise in image resolution, contrast, and frame rate. This thesis aims to enhance image quality for ultrafast ultrasound imaging using diverging waves, maintaining high frame rate and real-time performance. In particular,I first derived a closed-form approach that maps, under suitable hypotheses, the transmit apodization weights used in synthetic aperture imaging into the compound mask applied to diverging wave imaging. The approach draws inspiration from a successful technique developed for plane wave imaging, leveraging synthetic aperture imaging as a reference due to its superior image quality. Unlike the previous work on plane waves, the proposed approach is not limited to linear probe geometries but also works seamlessly with convex ones, thus expanding the scope of applicability of diverging wave imaging. Moreover, it handles arbitrary spatial arrangements of virtual sources generating divergent waves. The approach has been validated through simulated data using both linear and convex probes. After that, the weights are integrated into the beamforming pipeline of the Verasonics scanner, validating its efficacy across three deterministic VS configurations (linear, curvilinear, and tilted distributions). Experimental results demonstrate that the compound mask improves the quality of B-mode images with all distributions of virtual sources for linear and convex arrays, all without compromising real-time performance. Since the number and spatial distribution of these VSs affect both image quality and frame rate, their optimization is of high interest. I propose a multi-objective genetic algorithm to optimize VS spatial distributions with a compound mask weighting strategy to enhance beam coherence and reduce artifacts during optimization to further improve the image quality. The framework was evaluated across different numbers of VSs to quantify performance trade-offs under fewer transmission events. In both simulations and experimental trials, the proposed approach achieved improved image quality metrics against deterministic methods, preserving these gains even with reduced transmission events. Next, I developed a computationally efficient receive beamformer and coherent com pounder, suited to work with both radio frequency band and base band data related to plane-wave/diverging-wave insonifications under the constraints of a commercial ultrasound scanner, using GPU acceleration. The beamformer, implemented in GPU and taking into account the constraints of typical commercial ultrasound scanners, integrates multiple op timizations (precomputed delays, transmit/receive decoupling, symmetry exploitation, and carrier phase reconstruction ) to drastically reduce computation and data transfer overhead. Moreover, it integrates the optimization strategies previously developed (compound mask, optimized VS). The GPU beamformer is validated using in-vitro datasets collected from a tissue-mimicking phantoms, using a Verasonics Vantage system with a convex probe. Tests are performed with deterministic and optimized virtual source distributions and multiple VS configurations, including 20/10/6/4 transmissions. The proposed GPU-based receive beamformer achieves real-time performance while maintaining image quality even when the number of virtual sources is significantly reduced. Finally, the compound-mask apodization scheme generalizes from 2-D to fully volumetric ultrafast diverging-wave imaging with 2-D matrix probes, providing a practical basis for next-generation 3-D ultrasound. I devise a geometric, voxel-wise mapping that converts the apodization derived in 3-D synthetic transmit aperture imaging (STAI) into 3-D compound weights: for each virtual source and voxel, the source–voxel ray is traced to its intersection on the array face, yielding spatially consistent weights across the entire volume. The approach was assessed with FIELD II simulations, using matrix probes with full and randomly sparse elements and consistently reduce the sidelobe levels across ,depths.