Three-dimensional image reconstruction in breast ultrasound computed tomography

Li, Fu

Permalink

https://hdl.handle.net/2142/129556 Copy

Description

Title

Three-dimensional image reconstruction in breast ultrasound computed tomography

Author(s)

Li, Fu

Issue Date

2025-04-22

Director of Research (if dissertation) or Advisor (if thesis)

Anastasio, Mark A

Doctoral Committee Chair(s)

Anastasio, Mark A

Committee Member(s)

• Oelze, Michael
• Song, Pengfei
• Villa, Umberto

Department of Study

Bioengineering

Discipline

Bioengineering

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

Ultrasound tomography, breast imaging

Abstract

Ultrasound computed tomography (USCT) is an emerging imaging technique that uses tomographic principles to obtain quantitative estimates of acoustic properties such as speed-of-sound (SOS), density, and acoustic attenuation (AA). Because it can produce high-resolution and high contrast images of tissue properties, the development of USCT as a breast imaging modality has received significant attention. It has several advantages over other breast imaging modalities, such as mammography, including low cost and being radiation- and breast-compression-free. While commercial systems for breast USCT are being actively developed, USCT remains an emerging technology and a topic of active research. USCT systems commonly utilize a circular ring-array of elevation-focused ultrasonic transducers. Volumetric imaging is then achieved by translating the ring-array orthogonally to the imaging plane. Recent advancements in breast USCT employing such ring-array systems have demonstrated promising progress. However, image quality remains limited due to the use of simplified reconstruction methods that rely on a two-dimensional (2D) wave physics model. In this 2D approach, the three-dimensional (3D) wave propagation physics and the focusing properties of the transducers are not considered, resulting in images with significant artifacts and degraded spatial resolution. Therefore, developing advanced reconstruction algorithms that account for 3D wave propagation effects is important to further enhance breast USCT imaging. To address these challenges and advance 3D breast USCT imaging, this dissertation investigates two main aims. The first aim is to develop a virtual imaging framework for realistic \textit{in silico} 3D breast USCT imaging studies, including stochastic, realistic virtual breast phantoms and high-fidelity virtual data acquisition models. The second aim focuses on proposing advanced high-resolution 3D image reconstruction techniques, including model-based and deep learning-accelerated approaches. To advance the development and evaluation of reconstruction algorithms for new medical imaging technologies, computer simulation studies, commonly known as virtual imaging trials (VITs), are widely employed. VITs provide researchers with an ethical and controlled mean to explore imaging system designs and reconstruction methods, particularly for emerging technologies like USCT, which are still in the early stages of development. In VITs, it is essential to account for variability in the ensemble of objects to be imaged. This variability facilitates the assessment of task-based image quality (IQ) and the optimization of imaging system parameters. To this end, a methodology for generating realistic stochastic 3D numerical breast phantoms is developed, enabling clinically relevant computer simulation studies of USCT breast imaging. These phantoms incorporate anatomical and property variability representative of clinical settings, including differences in breast size, shape, composition, anatomy, and tissue properties. Next, as part of the virtual imaging framework, a high-fidelity 3D ring-array USCT data acquisition model is developed. The ring-array USCT system employs an elevation-focused ultrasonic transducer. Previous work often used simplified imaging models based on 2D wave physics. While computationally efficient, such models have several limitations, including mismatches between 2D and 3D wave physics and simplified transducer models, which are often assumed to be point-like. These mismatches eventually result in image artifacts and reduced spatial resolution in the reconstructed images. To address these challenges, a 3D USCT forward model was developed, incorporating the spatial impulse response of the transducer for accurate wave simulation. This model accounts for out-of-plane acoustic scattering and the transducer's focusing properties in both transmit and receive modes, which are crucial for enabling accurate ring-array USCT simulation and reconstruction. To overcome the limitations of the commonly used 2D SBS method and enable high-resolution, accurate USCT imaging, a 3D full waveform inversion (FWI) method is developed. This method incorporates 3D wave physics and transducer focusing properties. Additionally, a multi-ring 3D FWI method is implemented to further enhance reconstruction accuracy by utilizing acoustic data from multiple vertical transducer locations. The proposed 3D image reconstruction methods were evaluated through computer-simulation studies using the developed NBPs and clinical data. The impact of the number of ring-array positions on image accuracy and vertical resolution was also systematically assessed. Although 3D FWI has great potential for accurate, high-resolution imaging, its practical application in USCT has been limited by the significant computational burden associated with repeatedly solving the 3D acoustic wave equation. Thus, there remains an important need for algorithmic innovations that can accelerate 3D FWI so that image reconstruction times can be significantly reduced. To address this, a learning-based approach is developed to map the 3D ring-array USCT data to idealized 2D USCT measurements, allowing a faster and more accurate 2D reconstruction method. Additionally, the use of multiple ring-array measurements from adjacent elevations is explored as a multi-channel input to the neural network, showing improved accuracy compared to only using single-ring data. In summary, the methods proposed in this dissertation have the potential to push the boundaries of what a ring-array USCT system in transmit mode can achieve, enabling the creation of images that depict the distribution of acoustic properties in tissue with high spatial resolution and accuracy. Eventually, these methods may benefit women by improving the early detection and diagnosis of breast conditions, especially for those with dense breast tissue.

Graduation Semester

2025-05

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129556

Copyright and License Information

Copyright 2025 Fu Li

Owning Collections