Statistical analysis of positron emission tomography data

Thumbnail Image
PhD Thesis Final-Tian.pdf(24.35 MB)
Full Text E-thesis
Mou, Tian
Journal Title
Journal ISSN
Volume Title
University College Cork
Published Version
Research Projects
Organizational Units
Journal Issue
Positron emission tomography (PET) is a noninvasive medical imaging tool that produces sequences of images describing the distribution of radiotracers in the object. PET images can be processed to evaluate functional, biochemical, and physiological parameters of interest in human body. However, images generated by PET are generally noisy, thereby complicating their geometric interpretation and affecting the precision. The use of physical models to simulate the performance of PET scanners is well established. Such techniques are particularly useful at the design stage as they allow alternative specifications to be examined. When a scanner is installed and begins to be used operationally, its actual performance may deviate somewhat from the predictions made at the design stage. Thus it is recommended that routine quality assurance (QA) measurements could be used to provide an operational understanding of scanning properties. While QA data are primarily used to evaluate sensitivity and bias patterns, there is a possibility to also make use of such data sets for a more refined understanding of the 3-D scanning properties. Therefore, a comprehensive understanding of the noise characteristics in PET images could lead to improvements in clinical decision making. The main goals of this thesis are to develop model-based approaches for describing and evaluating the statistical properties of noise and a practical approach for simulation of an operational PET scanner. We began with the empirical analysis of statistical characteristics—bias, variance and correlation patterns in a series of operational scanning data. A multiplicative Gamma model had been developed for representing the structure of reconstructed PET data. The novel iteratively re-weighted least squares (IRLS) techniques were proposed for the model fitting. These included the use of a Gamma-based probability transform for normalising residuals, which could be used for model diagnostics. Building on the Gamma based modelling and probability transformation, we developed a 3-D spatial autoregressive (SAR) model to represent the 3-D spatial auto-covariance structure within the normalised data. Auto-regressive coefficients were also estimated based on the minimisation of difference between 3-D auto-correlations calculated from the normalised data and model. Both traditional filtered back-projection (FBP) and expectation-maximisation (EM) reconstructions were considered. Numerical simulation studies were carried out to evaluate the performance of the above models. The proposed models led to a very trivial process for simulation of the scanner—one that can be implemented in R. This provided a very practical mechanism to be routinely used in clinical practice—assessing error characteristics associated with quantified PET measures. Moreover, this fast and simplified approach has a potential usage in enhancing the quality of inferences produced from operational clinical PET scanners.
PET , Image processing , Gamma distribution , IRLS , Spatial autocorrelation , Simulation , Iterative EM reconstruction , Conditional Likelihood , Quality assurance
Mou, T. 2018. Statistical analysis of positron emission tomography data. PhD Thesis, University College Cork.