Kinetic Model Based Factor Analysis of Cardiac Rubidium-82 PET Images for Improved Accuracy of Quantitative Myocardial Blood Flow Measurement

Coronary artery disease is a leading cause of death, is routinely diagnosed using myocardial perfusion imaging (MPI), and can be managed effectively with proper therapy. However, uniform reduction in flow throughout the heart due to disease in multiple arteries may not be detected with MPI. Myocardial blood flow (MBF) quantification using positron emission tomography (PET) can overcome this limitation, but has limited clinical application due to a need for an onsite cyclotron. 82Rb PET MPI does not require a cyclotron and is being applied widely.;Factor analysis (FA) can decompose dynamic images into underlying components of the image, but requires constraints to ensure physiological accuracy. A model-based FA method (MB) that incorporates the tracer kinetic model into the FA process as a constraint is developed and compared with a previously proposed minimal-structure-overlap FA method (MSO). In simulations, MB was more accurate and reproducible than MSO. In rat experiments with arterial blood sampling as a standard, MB resolves more physiologically accurate blood TACs. Structures were more reproducible with MB vs. MSO in repeat images of the same dog with variable-length 82Rb infusion durations, and MBF estimates tended to be more reproducible.;The accuracy of MBF in humans using ROI-based and MB-based methods was evaluated using 15O-water imaging as a standard, but no significant differences were found. However, MBF regional uniformity in normals was significantly improved over ROI based methods. In a patient population uniformity was not significantly different between methods, indicating that uniformity was not artificial. Thus MB based MBF values may be more sensitive to detect small changes in MBF.;In this work, a region-of-interest (ROI) based method to quantify MBF from dynamic 82Rb PET images was developed. Blood and myocardium time-activity-curves (TACs) were generated from dynamic PET images and used as input and output functions respectively to a tracer kinetic model. MBF was resolved by fitting the model to the TACs. The highly automated method had little operator-dependent variability of MBF. However, due to the limited resolution of PET, signal from myocardial tissue can spillover into blood regions, contaminate the blood TACs, and can degrade the accuracy of MBF.