The goal of the proposed research is to develop computationally efficient image-reconstruction methods to improve the quality of cardiac-perfusion image sequences obtained by gated single-photon emission computed tomography (SPECT). Methods for reconstruction of a single, static tomographic image have been thoroughly investigated over the past 20 years, but relatively little attention has been paid to the specific problem of image-sequence reconstruction. Currently, image sequences are computed, one image frame at a time, by methods designed for the reconstruction of static images. This approach is highly suboptimal because it fails to take into account the strong statistical correlations that exist among the image frames. We propose that image quality can be substantially improved by treating an image sequence, not as a set of individual frames, but as a single four-dimensional (4D) signal, described by the three spatial dimensions, plus time. We expect this approach to produce cardiac perfusion images that are more accurate, and less noisy, than those obtained currently. In the proposed project, a collection of new, 4D image-reconstruction techniques will be developed, implemented, and evaluated. The objective of the research is to improve the image quality achieved by gated cardiac SPECT, and thus improve its accuracy in the detection and evaluation of coronary artery disease. The results will be evaluated by using measures of diagnostic task performance based on human and machine observers. Specifically, improvements will be judged based on performance measures related to perfusion defect detection, wall motion, wall thickening, and ejection fraction. Although the research will focus on cardiac SPECT, the techniques developed are expected to be applicable to other nuclear medicine applications, and image-sequence processing for other imaging modalities.
Showing the most recent 10 out of 20 publications
