Date of Award
12-20-2024
Degree Name
Doctor of Philosophy (PhD)
Degree Type
Dissertation
Abstract
Attenuation compensation (AC) is beneficial for visual interpretation and serves as a prerequisite for quantification tasks in single-photon emission computed tomography (SPECT) imaging. However, conventional AC methods typically rely on a separate X-ray CT component, leading to multiple challenges. In this context, studies have shown that scatter-window projection data contains information to estimate the attenuation distribution. Furthermore, large amounts of SPECT emission data in both photopeak and scatter-energy windows and corresponding CT scans are available. The ability of deep learning (DL) to model complex relationships by leveraging large datasets motivates its use for transmission-less AC in SPECT. Building upon the idea of integrating physical and DL techniques, this dissertation proposes scatter-window projection and DL-based AC methods for SPECT: CTLESS for myocardial perfusion SPECT and DaT-CTLESS for dopamine transporter SPECT. The proposed CTLESS method was evaluated on the clinical task of cardiac perfusion defect detection. This task-specific evaluation was motivated by another study conducted in this dissertation demonstrating that evaluating using visual fidelity metrics may not correlate with performance on clinical tasks. We evaluated the CTLESS method on the clinical task in both an anthropomorphic model study and a multi-reader multi-case human observer study involving physician readers with expertise in detecting cardiac defects in myocardial perfusion SPECT images. The CTLESS method yielded similar receiver operating characteristic (ROC) curves and area under the ROC curve (AUC) compared to the standard-of-care CT-based AC method (CTAC) and demonstrated statistical non-inferiority to CTAC. Moreover, CTLESS significantly outperformed a method without AC (NAC). These findings were consistent across stratified analyses by sex and defect types. Additionally, CTLESS demonstrated strong generalizability on the clinical task across different SPECT scanners. The proposed DaT-CTLESS method was evaluated in an \textit{in silico} imaging trial that simulated patient variability and SPECT system physics. DaT-CTLESS yielded a similar performance to CTAC and significantly outperformed an AC method that uses uniform attenuation maps (UAC) on the task of regional uptake quantification. Furthermore, DaT-CTLESS significantly outperformed UAC in distinguishing patients with normal versus reduced putamen-specific binding ratios. These findings demonstrate the capability of CTLESS and DaT-CTLESS for transmission-less AC in SPECT and provide evidence for the potential clinical translation.
Language
English (en)
Chair
Abhinav Jha
Committee Members
Barry Siegel; Farhan Katchi; Quing Zhu; Yuan-Chuan Tai