A 4D Statistical Atlas of the Human Heart from Gated PET Myocardial Perfusion Imaging

Objectives: PET myocardial perfusion imaging (MPI) has emerged as an excellent imaging modality to diagnose coronary artery disease and provide prognostic information in a wide range of pathologies. Although PET MPI allows absolute quantification of myocardial blood flow and quantitative assessment of left ventricular ejection fraction, regional wall motion analysis is primarily performed qualitatively. The ability to automate quantitative assessment of regional wall motion could provide additional diagnostic and prognostic information. We are developing methodology for a normative, intersubject, and spatiotemporal four-dimensional (4D) atlas of the left ventricle to support statistical comparison and quantitative evaluation of the characteristics of human cardiac function. The goal of this work is to report a 4D atlas of the human heart using gated PET MPI in which normal intersubject variability is summarized in the voxelwise mean and variance of the functional endpoints across various subjects. Statistics of the myocardial motion pattern in a healthy population can be succinctly summarized by the principle components (PCs) of an orthogonal decomposition of the 4D atlas.

Methods: Twelve healthy human subjects were imaged on a Siemens Biograph PET/CT system using 13N-NH3. The scan included a standard stress acquisition with the single-scan, two-injection method that was previously proposed and validated^[1-3]. The acquisition is performed at peak stress with an adenosine infusion (140 mcg/kg/min) and 13N-NH3 injection (600 MBq). Gated PET image volumes were dynamically reconstructed with 16 gates per cardiac cycle. To define a basic structural framework for atlas construction, we first created a common physical space using a diffeomorphic groupwise registration method. The gated PET volumes of all subjects were transformed into this common space using the deformations fields computed simultaneously during the registration process. At each of the 16 time bins, the deformed PET volumes were fused together, yielding a gated 4D PET atlas. The mean and variance of the fused volumes were computed to summarize the statistical properties of the atlas. In addition, we incorporated a principal component analysis (PCA) framework to characterize myocardial motion. Consecutive diffeomorphic registrations from the end-diastole (ED) as reference phase to the end-systole (ES) phase are compared. The variability of the myocardial motion within a population are reflected in the weights and directions of the PCs.

Results: The proposed 4D PET atlas was successfully constructed with high quality mean intensity image and a standard deviation image describing intersubject variability (Figure 1(a)). Specifically, statistics of the mean and standard deviation images at ED and ES are listed in Table 1. Subject-specific myocardial motion was computed at ES phase. The motion field and PC directions were visualized (Figure 1(b)) and the weight percentages of the PCs are shown in Table 2. In general, PC #1 follows the myocardium circumferentially and PC #3 follows it vertically.

Conclusions: The 4D gated cardiac PET atlas provides myocardial images of a standard size and orientation combining anatomic and physiological data. This is the first atlas combining anatomic and functional images in a standard space. This provides a tool to compare the differences and variability of a variety of parameters in a population and allows us to analyze the variability of the myocardial anatomies and cardiac wall motion. This method can potentially be extended to clinical applications involving automated image interpretation. Acknowledgments: This work is supported by NIH P41EB022544 and R01HL137230. We thank Marina T. Macdonald-Soccorso for her effort on PET MPI reconstruction.

• Download figure
• Open in new tab
• Download powerpoint