Data-driven method for parametric imaging using dynamic cardiac single photon emission computed tomography

Dynamic cardiac SPECT imaging is a technique developed to measure as well as model the kinetics of a tracer in the myocardium. The kinetic rate parameters obtained represent the absolute physiological measures, thereby providing information for objective analysis. The use of fast temporal sampling and the selection of small regions of interest, however, result in high statistical noise in the time-activity curves causing inaccuracies in the kinetic rate parameters. In order to obtain reliable parametric information through the use of less noisy time-activity curves, a data-driven clustering-based segmentation method is developed for dynamic cardiac SPECT imaging.; A comparative study of dynamic and static Thallium-201 SPECT perfusion imaging was performed in canines and patients. In order to suitably segment the dynamic data, a temporal clustering algorithm was developed to group similar time-activity curves. The information of the blood input function determining blood cluster was used for the selection of the best number of clusters for any given study. A preliminary method combining spatial and temporal information in the clustering algorithm was designed. Further, a study of the different clustering parameters as a function of number of clusters and weighting factor was performed. Computer simulations as well as experimental studies in canines and patients were performed using Thallium-201 and 99mTechnitium-labeled teboroxime to validate the methods.; The wash-in rate parameters obtained with dynamic Thallium-201 SPECT gave a higher and more accurate defect contrast than those obtained with static SPECT. The defect size was overestimated in some cases, likely due to the use of a suboptimal revised temporal sampling scheme consisting of fewer longer samples. When temporal clustering was used to segment dynamic Thallium-201 data, the wash-in rate estimates obtained gave better correlation with the microsphere derived flows compared to those obtained with the semiautomatic method. For dynamic 99mTechnitium-labeled teboroxime canine studies, spatio-temporal clustering method gave wash-in rate estimates that were slightly closer to the microsphere-derived flows than those obtained with temporal clustering.; Dynamic cardiac SPECT imaging along with the data-driven clustering method has the potential to provide more accurate myocardial perfusion information than static imaging, thereby improving the diagnosis and management of coronary artery disease.