Adaptive Imaging and SPECT

Abstract

Single-photon emission computed tomography (SPECT) systems are widely used in the field of medical imaging as a tool to perform a variety of imaging tasks, ranging from basic biological studies to pre-clinical drug development. However, different imaging tasks require different types of images---for example, one task may benefit from images with high contrast and another task may prefer images with low noise. This leads to a multitude of SPECT systems, each having a unique set of imaging parameters that have been specifically chosen to generate good images based on the task. Computer simulations are used to model and test different imaging configurations to determine the optimal imaging parameters. This optimization task, however, becomes more challenging when the number of possible configurations becomes very large. In this dissertation, we present a collection of methods that can be used to model and optimize SPECT systems with a large parameter space. Two imaging systems are examined in detail: the synthetic collimator SPECT system and an adaptive imager known as AdaptiSPECT. A task-based approach is used to provide an objective measure of the imaging performance. Under this framework, system configurations are defined as being better based on how well a particular imaging task is accomplished by using the images generated, rather than other image metrics like resolution and contrast. The synthetic-collimator SPECT system is used to estimate the uptake of dopamine in different mouse brain regions. Multi-pinhole apertures are used to generate images with both high-resolution and high-sensitivity. However, multi-pinhole apertures can also introduce image multiplexing, which has been shown to result in artifacts in reconstructions, possibly degrading imaging performance. To mitigate these effects, a pair of semi-conductor detectors are used to simultaneously acquire two images of the object at the same projection angle, but at different magnifications. The data acquired by the front detector have low levels of multiplexing and can be used to resolve some of the uncertainty in the data collected by the rear detector, which suffers from greater multiplexing. Different system configurations are constructed by varying the two detector distances as well as the separation between the multiple pinholes. The Wiener estimator is used to examine the performance of each configuration. Because the Wiener estimator only requires the first- and second-order statistics of the object ensemble and projection images, we are able to efficiently and quickly scan through the possible parameter configurations to determine good imaging configurations. We also demonstrate that the optimal detector distances of the synthetic-collimator SPECT system are closely related to the amount of multiplexing present on each detector. The AdaptiSPECT system is a state-of-the-art imager that is capable of changing its imaging parameters in real-time during the imaging acquisition. The system consists of 16 NaI(Tl) gamma cameras which are allowed to move independently of one another, changing their distance relative to the system's aperture. The aperture also has several adjustable components, allowing for variations to the pinhole radius and aperture distance, measured relative to the center of the field-of-view. The aperture can also switch between single- and multi-pinhole configurations. Different configurations to the AdaptiSPECT system are uniquely defined by a system operator and the process of generating the system operator is discussed in detail. The ability to change the parameters of AdaptiSPECT make the system uniquely suitable for adaptive imaging. Adaptive imaging is the process of customizing the imaging parameters of a system based on the unique properties of objects, in an effort to improve imaging performance. The imaging task of interest is the estimation of the strength of a spherical signal located in a noisy background. Our adaptive strategy consists of two imaging scans: a scout scan and a diagnostic scan. Image data acquired from the scout scan are reconstructed to estimate object features such as object support and distribution of activity. The object features are then used to determine the optimal system parameters for the diagnostic scan. The diagnostic scan collects data from both single- and multi-pinhole configurations of AdaptiSPECT and the combined dataset is reconstructed by using the Alternating List-mode Maximum-Likelihood Expectation-Maximization algorithm (A-LMMLEM). A region-of-interest estimator is then used to estimate the signal strength. We show that an adaptive system using our optimization algorithm outperforms a system with fixed parameters. While the algorithm does not identify the best possible system configuration, it can be performed in real-time and results in system configurations that perform close to the best system, without the need of an exhaustive search.