TY - JOUR
T1 - <strong>Direct Bayesian parametric image reconstruction from dynamic myocardial perfusion PET data</strong>
JF - Journal of Nuclear Medicine
JO - J Nucl Med
SP - 1772
LP - 1772
VL - 59
IS - supplement 1
AU - Li, Yang
AU - He, Yuru
AU - Rahmim, Arman
AU - Lu, Lijun
Y1 - 2018/05/01
UR - http://jnm.snmjournals.org/content/59/supplement_1/1772.abstract
N2 - 1772Objectives: Dynamic myocardial perfusion PET imaging requires application of kinetic modeling to time activity curves (TACs) generated at the voxel level, which can result in poor quantification of myocardial blood flow (MBF) maps. To improve quantitative accuracy of kinetic parameters at the voxel level, a direct Bayesian parametric image reconstruction framework was developed. Methods: Conventional indirect parametric reconstruction (i.e. fitting voxel-wise TACs) assumes that the noise in the image domain is Gaussian distributed and neglects correlation between voxels, resulting in unacceptable noise levels and quantitative performance. The direct 4D spatiotemporal method combines kinetic modeling and emission image reconstruction into a single framework, enabling accurate modeling of noise from sinogram data into the reconstruction. We developed a direct maximum a posteriori (MAP) parametric image reconstruction framework including a Poisson log-likelihood function and a quadratic penalty function. The objective function was solved by a preconditioned conjugate gradient (PCG) algorithm with a developed preconditioner whose entries are the ratios between the kinetic parameters and the sensitivity of the expected detected counts due to them respectively to optimize the convergence rate of iteration. Using simulated dynamic myocardial perfusion Rb-82 PET data, the initial parametric image was first estimated from the conventional indirect method with a weighted non-linear least-square fitting algorithm. Then the objective was optimized by the developed PCG algorithm from the initial parametric image, which is assumed to be in the same concave as the ground truth. Result: The proposed direct MAP reconstruction can converge to a robust solution. The estimated parametric image of K1 from the direct method produced lower standard deviation (lower by 52.3%) relative to the indirect method, for matched mean value of K1was almost the same (lower by 0.46%). Conclusions: The proposed direct MAP method generated visible improvement at the matched bias level and accuracy quantification compared with indirect Methods: Acknowledgments: This work was supported by the National Natural Science Foundation of China under grants 61628105, 81501541, the National key research and development program under grant 2016YFC0104003, the Natural Science Foundation of Guangdong Province under grants 2016A030313577, and the Program of Pearl River Young Talents of Science and Technology in Guangzhou under grant 201610010011.
ER -