Quantitative reconstruction for brain SPECT with fan-beam collimators

SPECT (single photon emission computed tomography) is well known for its providing functional information of fairly good quality and its relatively low cost in diagnosis. It is one of the most important modalities to demonstrate the blood flow, the perfusion, and oxygen and glucose metabolic activities. However, there are several inherent factors in current available systems that degrade the quality of reconstructed images, among which the absorption of γ-rays by the body and the non-stationary detector blurring effect due to collimation/scintillation process are two major ones. Non-parallel collimations such as fan-beam geometry designed to improve the detection resolution and sensitivity makes the image reconstruction even more complicated. To date, the methods to deal with both attenuation and detector blurring effects are based on iterative algorithms, which are approximated and time consuming. Our aim is to find a more efficient and accurate method to solve this problem in a specific application—brain SPECT.; The described reconstruction procedure has been shown to be an optimal solution to the problem of fan-beam brain SPECT imaging in terms of both the computational efficiency and the accuracy. It has been validated by the mathematical disks phantom, Shepp-Logan phantom and Hoffman brain phantom using Monte Carlo simulations. Results showed satisfactory reconstructions of the phantoms, and the computation time is less than 10 minutes for a 128 x 128 x 64 volumetric image on a Pentium III 550MHz PC platform. The method is compared with two other methods, one uses the frequency distance relation (FDR) for deconvolving the PSF blurring effect after the same noise and scatter treatment, then uses exponential weighted filtered backprojection method to reconstruct the uniformly attenuated projection; the other uses the ordered subset expectation maximization (OSEM) iterative algorithm with all degradation factor accurately modeled inside. The comparisons show that the CHD-CG method along with accurate noise modeling and scatter compensation could achieve the same image quality as the OSEM iterative method with improved computation efficiency. The OSEM reconstruction for the same image volume on a same PC platform needs more than 8 hours. Quantitative analysis on regional bias-variance, ROC (receiver operating characteristic) study, and Hotelling trace calculation are also evaluated for these methods, which show comparable performance between the proposed method and the OSEM iterative method in term of defect detectability. Most of the images in this dissertation are generated by C/C++ program under Windows system, other plots are calculated with MATLAB, and the Monte Carlo simulation uses FORTRAN codes running under UNIX system. (Abstract shortened by UMI.)...