Spectral ct at ultra-high resolution via photon-counting and deep learning

Li, Mengzhou
Thumbnail Image
Other Contributors
Hahn, Juergen
Yan, Pingkun
Lai, Rongjie
De Man, Bruno
Wang, Ge
Issue Date
Biomedical engineering
Terms of Use
This electronic version is a licensed copy owned by Rensselaer Polytechnic Institute (RPI), Troy, NY. Copyright of original work retained by author.
Full Citation
Computed tomography (CT) plays an indispensable role in the clinical diagnosis and treatment of diseases. Enabled by recent dual-energy CT (DECT) advances, spectral CT allows anatomical and functional imaging with material characterization capabilities. Despite the great utilities, current CT technologies still suffer from insufficient spatial resolution and spectral fidelity for important clinical applications, especially in cardiovascular examination and temporal bone imaging. A breakthrough in high-fidelity spectral imaging at ultrahigh resolution (on the order of 50µm) would greatly benefit otology & neurotology, cardiology and other important clinical applications. The emerging X-ray photon-counting detectors (PCDs) make such imaging possible, with their incredibly small detector elements and impressive photon energy discrimination ability. Despite their promising potentials for medical CT, there are still major obstacles for clinical translation: (1) the imperfectness of current PCDs may cause severe spectral distortions by charge-sharing and pulse pileup effects, and hinder the fidelity of spectral imaging and limit the spatial resolution; (2) to keep image quality at a similar level for finer resolution, the radiation dose needed is proportional to the fourth power of the resolution increase which could be a huge concern; and (3) with drastically improved system resolution, the sensitivity to patient motion and geometry misalignment becomes more prominent and can be the bottleneck limiting the practical resolution. To overcome these challenges, our overall goal is to develop cutting-edge techniques and algorithms empowered by deep learning to pave the way for photon-counting spectral CT at an ultra-high spatial resolution to enter clinical practice, and come up with a clinical micro-CT (CMCT) prototype design with an initial emphasis on temporal bone ultra-high resolution imaging. Towards the goal, this thesis is organized around following specific aims targeting aforementioned challenges: (1) develop a deep learning-based photon counting data calibration approach for high-fidelity spectral imaging; (2) develop and optimize advanced geometric calibration, motion correction, interior tomography, and dose-saving image reconstruction methods dedicated for high-quality ultra-high resolution imaging at minimized radiation dose; (3) design and prototype a CMCT system with simulation tools according to clinical needs for temporal bone imaging (∼ 50µm), and demonstrating the clinical feasibility and imaging capabilities in phantoms or animal models both numerically and physically. A set of techniques have been developed with simulation and validated on real data upon achieving these aims, which will be a giant leap forward pushing the dream of CMCT into reality.
School of Engineering
Dept. of Biomedical Engineering
Rensselaer Polytechnic Institute, Troy, NY
Rensselaer Theses and Dissertations Online Collection
Restricted to current Rensselaer faculty, staff and students in accordance with the Rensselaer Standard license. Access inquiries may be directed to the Rensselaer Libraries.