Doctor of Philosophy

dissertationSingle Photon Emission Computed Tomography (SPECT) myocardial perfusion imaging (MPI), a noninvasive and effective method for diagnosing coronary artery disease (CAD), is the most commonly performed SPECT procedure. Hence, it is not surprising that there is a tremendous market need for dedicated cardiac SPECT scanners. In this dissertation, a novel dedicated stationary cardiac SPECT system that using a segmented-parallel-hole collimator is investigated in detail. This stationary SPECT system can acquire true dynamic SPECT images and is inexpensive to build. A segmented-parallel-hole collimator was designed to fit the existing general-purpose SPECT cameras without any mechanical modifications of the scanner while providing higher detection sensitivity. With a segmented-parallel-hole collimator, each detector was segmented to seven sub-detector regions, providing seven projections simultaneously. Fourteen view-angles over 180 degree were obtained in total with two detectors positioned at 90 degree apart. The whole system was able to provide an approximate 34-fold gain in sensitivity over the conventional single-head SPECT system. The potential drawbacks of the stationary cardiac SPECT system are data truncation from small field of view (FOV) and limited number of view angles. A tailored maximum-likelihood expectation-maximization (ML-EM) algorithm was derived for reconstruction of truncated projections with few view angles. The artifacts caused by truncation and insufficient number of views were suppressed by reducing the image updating step sizes of the pixels outside the FOV. The performance of the tailored ML-EM algorithm was verified by computer simulations and phantom experiments. Compared with the conventional ML-EM algorithm, the tailored ML-EM algorithm successfully suppresses the streak artifacts outside the FOV and reduces the distortion inside the FOV. At 10 views, the tailored ML-EM algorithm has a much lower mean squared error (MSE) and higher relative contrast. In addition, special attention was given to handle the zero-valued projections in the image reconstruction. There are two categories of zero values in the projection data: one is outside the boundary of the object and the other is inside the object region, which is caused by count starvation. A positive weighting factor c was introduced to the ML-EM algorithm. By setting c>1 for zero values outside the projection, the boundary in the image is well preserved even at extremely low iterations. The black lines, caused by the zero values inside the object region, are completely removed by setting 0< c