Temporal regularization and artifact correction in single slow-rotation dynamic SPECT

Single photon emission computed tomography (SPECT) is a diagnostic functional imaging modality wherein the distribution of a radioactive tracer inside the body is estimated based on data acquired from around the patient by a slowly rotating camera. Conventional SPECT image reconstruction assumes that this distribution remains constant during acquisition. In this thesis we investigate imaging of a time-varying distribution of radiotracer, which results in a highly underdetermined reconstruction problem. Recovery of an accurate dynamic image from this data requires the use of additional constraints, including temporal regularization. This work builds on the dSPECT approach of Farncombe et al., which uses simple inequality constraints to restrict the temporal behaviour of the reconstructed image. We first investigate the use of a stronger temporal constraint than the one used in dSPECT, to improve the quality of reconstructed images. Since dynamic tracer behaviour in the human body arises as a result of continuous physiological processes, changes in tracer concentration should follow a smooth time activity curve (TAC). We propose a modification to dSPECT, denoted d2EM, which promotes smoothness by constraining the concavity of the TAC in every voxel of the reconstructed image. Digital phantom simulations comparing the performance of d2EM versus the original dSPECT algorithm show that d2EM provides more accurate images, with smoother, more consistent TACs within dynamic regions of interest. The d2EM method is especially successful in simulations featuring high levels of noise and relatively gradual tracer kinetics. We then examine artifacts in dynamic images reconstructed from single slow-rotation data, which occur due to the fact that only a small number of views of the object are acquired by the camera at any one time. Particular emphasis is placed on image artifacts caused by the effects of attenuation on the projection data, which can be severe. Using realistic 3D phantom simulations, as well as real-life dynamic renal SPECT data, we investigate methods to correct for these artifacts. The correction methods are shown to substantially improve the accuracy of reconstructed TACs in the presence of attenuation.