Statistical modeling of radiation detection and imaging system

In the development of SPECT system for imaging small animal in vivo, higher spatial resolution is continuously sought to provide improved spatial detail and image quality. However, the pursuit of high resolution often results in the poor photon collection efficiency and limited quantitative accuracy, which ultimately confines the capacities of the imaging modality. Adaptive imaging approach could be implemented in order to maximize the efficiency for collecting useful imaging information regarding a given task and therefore provide an optimum image performance.;In the adaptive data acquisition, the system hardware or acquisition protocol could vary in response to information being acquired during an imaging study, which requires a reasonable optimization method to conduct the system configuration alteration. In practical system optimization process, the key challenge is to optimize the system performance in real-time with respect to a wide range of design and imaging parameters for observing the unknown object. Even with state-of-the-art parallel computing platform, we still have to face two main difficulties: large computation load for performance indices evaluation and the complexity of the optimization problem against different system parameters. In order to address these issues, we have developed a series of approaches to enable SPECT system optimization with reasonable computation load.;As the first step towards the system optimization, we have developed a vector modified uniform Cramer-Rao bound (MUCRB) to replace the time-consuming brute-force MC simulation. This approach allows one to analytically derive the fundamental tradeoffs between resolution and minimum achievable total (or average) variance over arbitrarily chosen voxels, which could be asymptotically achieved by the post-filtered penalized maximum likelihood estimation with well-defined penalty function in linear Poisson model. In order to further reduce the computation load, the non-uniform object-space pixelation (NUOP) approach has been developed to divide the object-space into smaller voxels for target-regions, and into larger voxels in areas that are relatively smooth and/or less important to the reconstruction of the target-regions. This method could improve the calculation speed for reconstruction and vector MUCRB calculation by 1-2 orders of magnitude without sacrificing image quality inside the target-regions. The combination of these two approaches could allow real-time system performance evaluation.;Based on these techniques, we have also developed a generic system optimization method that allows the system to be optimized against any arbitrarily given system parameters. This approach was first used to develop an adaptive angular sampling approach for SPECT imaging. This approach allows the camera to spend larger fractions of imaging time at angles those are relatively more efficient for acquiring useful imaging information from the target to deliver significantly lowered image variance at a given resolution, which builds a foundation for further optimizing performance with respect to many other system parameters. We have further expanded this method to develop an indirect system optimization approach dealing with a discretized the parameter space. It helps to identify the best combination of different system configurations to be used for attaining the optimum imaging performance. The series of analytical approaches developed through this thesis work could be used together to provide an efficient computation scheme to facilitate real-time system optimization. It could be used along with variable hardware detection system for use in adaptive SPECT imaging.