4D Dynamic Image Reconstruction for Cardiac SPECT

• Main Author: Abdalah, Mahmoud
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01.01.2014
• Subjects: Computer science; Medical imaging
• ISBN: 1321552882; 9781321552881

Imaging methods in nuclear medicine such as Single Photon Emission Computed Tomography (SPECT) and Positron Emission Tomography (PET) attempt to grasp the functionality of imaged organs. Static SPECT aims to reconstruct 3D representation of the tracer's concentrations inside the imaged subject, while dynamic single photon emission computed tomography (dynamic SPECT) aims to estimate the 4D concentration distribution of a gamma-ray emitting tracer inside the patient as it changes with time. The temporal variation of the concentration function, represented by time activity curves (TACs), provides important information about radiotracer pharmacokinetics and about physiology of tissues and organs. Dynamic SPECT is the main focus of this research. Several issues are studied and addressed by this dissertation including the development of a dynamic reconstruction algorithm, a procedure for initializing the reconstruction algorithm, and a method for estimating the proper value of the regularization parameter. First, we have developed and validated an algorithm for extracting voxel-by-voxel time activity curves directly from inconsistent projections applied in dynamic cardiac SPECT. The algorithm derived was based on the factor analysis of dynamic structures (FADS) approach and imposes prior information by applying several regularization functions. The anatomical information of the imaged subject is used to apply the proposed regularization functions adaptively in the spatial domain. The algorithm performance is validated by reconstructing dynamic simulated datasets using the NCAT phantom with a range of different input tissue time-activity dependent curves. The validated algorithm is then applied to reconstruct pre-clinical cardiac SPECT data from canine, murine, and human subjects. Second, since dynamic SPECT is an underdetermined problem, initialization is important for the algorithm to converge to the right solution. Therefore, we have developed a procedure for initializing the algorithm using an initial reconstruction with a set of b-splines. In the reconstruction, we extract a set of initial time activity curves and use them for initialization. During the dynamic reconstruction, the algorithm imposes several regularization functions. The weights (also called regularization parameters or hyper-parameters) of those functions are estimated dynamically from the data by adapting a method that optimizes for those parameters and updates them at every iteration. With these improvements, we have successfully validated our method using numerically simulated data, and then applied it to three different types of real SPECT data acquired from slow gantry rotation. The results show that our developed algorithm is robust and has stable results.