Development of a Physics-consistent 3D Ultrasound Image-reconstruction Framework for High Fidelity Volumetric Imaging

Abstract

Three-dimensional （3D） ultrasound imaging can provide more complete spatial information than conventional two-dimensional imaging, but accurate volumetric reconstruction remains challenging due to limited acquisition geometry, simplified beamforming assumptions, and incomplete recovery of target structures. In many conventional methods, the propagation medium is assumed to have a constant speed of sound, and delay-and-sum （DAS） beamforming is implemented using a simplified discrete- delay approximation. These assumptions can lead to delay mismatch, degraded focusing, limited-view artifacts, and reduced volumetric fidelity. This thesis presents a simulation-based, physics-consistent framework for 3D ultrasound image reconstruction to achieve high-fidelity volumetric imaging. The proposed framework combines heterogeneous acoustic modeling, multi-transmit acquisition, fractional- delay DAS beamforming, and image-space compounding. Full-wave ultrasound simulation is used to generate radio-frequency （RF） data from controlled 3D phantom configurations, followed by preprocessing steps including time gating, bandpass filtering, and analytic signal formation. Volumetric reconstruction is then performed using fractional-delay DAS together with receive apodization and coherence-factor weighting. Finally, compounded 3D images are obtained through envelope detection, log compression, and visualization. A key contribution of this work is the emphasis on sound-speed- related physical consistency. Unlike simplified conventional approaches, the proposed framework models heterogeneous acoustic media in the forward simulation and establishes a variable-speed delay formulation for more realistic beamforming. In addition, fractional-delay DAS improves sub-sample delay alignment, while multi-transmit acquisition and image-space compounding improve spatial coverage, boundary continuity, and target visibility. The framework was evaluated using single-sphere, multi-sphere, and two-target resolution phantoms. The results showed that the baseline limited-view configuration produced incomplete sphere recovery and characteristic C-shaped artifacts. Multi- transmit compounding and parameter refinement improved target coverage, and the final integrated framework produced more complete boundary recovery, reduced artifacts, improved contrast, and better 3D consistency. Multi-sphere experiments confirmed positional consistency, while resolution analysis showed that two closely spaced targets became distinguishable at an approximate separation of 1.0 mm. Overall, this thesis demonstrates that the integration of physics-consistent modeling and advanced beamforming strategies can substantially improve 3D ultrasound volumetric reconstruction quality and provides a useful foundation for future real-data validation and sound-speed-aware reconstruction.