- Restrictions on Access:
- Restricted (PSU Only).
- Ultrasound Computed Tomography (UCT) is a non-invasive, inexpensive, radiation-free medical imaging technique that is able to resolve soft tissue structures in the body for advanced clinical diagnosis. Waveform inversion-based image reconstruction methods are frequently employed in UCT as they produce high-resolution images by accounting for second-order wave-propagation effects. Waveform inversion methods in UCT are often modeled as an optimization scheme solved by iterative methods and these iterative methods are resource-intensive. Though very successful, waveform inversion methods are computationally expensive and the optimization problem is non-linear, ill-posed, and struggles when dealing with high contrasts between tissue types without a good initial model. This can be attributed to the large difference in the contrast which requires solving a large-scale optimization scheme through iterative gradient-based processes. Thus, existing modalities in UCT are limited when reconstructing high contrast tissues/organs and far from real-time consideration. In this thesis, we propose to leverage the promise of Deep Learning (DL) to overcome the computational burden and ill-posedness to achieve near real-time image reconstruction in ultrasound tomography. We aim to directly learn the mapping from the recorded time-series sensor data to a spatial image of acoustical properties. This thesis discusses two DL-based methods for UCT; DeepUCT and DL-CRF. In DeepUCT, we developed a deep learning model using two cascaded Convolutional Neural Networks (CNN) with an encoder-decoder architecture. We achieve a good representation by first extracting the intermediate mapping knowledge and later utilizing this knowledge to reconstruct the image. Whereas in DL-CRF, the presence of high contrast phantoms is overcome by directly reconstructing the ultrasound data using a deep CNN network. The deep CNN network is then coupled with Conditional Random Fields (CRFs) to eliminate the artifacts introduced by the CNNs and improve the structural details in the reconstructed image. In both cases, the networks are developed using an encoder-decoder architecture which reconstructs acoustic property distribution of the recorded ultrasound data in a fraction of a second. These approaches are evaluated on synthetic phantoms where simulated ultrasound data are acquired from a ring of transducers surrounding the region of interest. The measurement data is acquired by forward modeling the wave equation using the k-wave toolbox. Our simulation results demonstrate that our proposed deep-learning method is robust to noise and significantly outperforms the state-of-the-art traditional iterative method both quantitatively and qualitatively. Furthermore, our model took substantially less computational time than the conventional full-wave inversion method.
- Dissertation Note:
- M.S. Pennsylvania State University 2021.
- Technical Details:
- The full text of the dissertation is available as an Adobe Acrobat .pdf file ; Adobe Acrobat Reader required to view the file.
View MARC record | catkey: 35144127