Quantitative brain PET imaging incorporating an anatomy-guided joint prior model

Objectives To improve quantitative accuracy of brain PET imaging through a novel anatomy-guided joint prior.

Methods We proposed a maximum a posteriori (MAP) framework incorporating information from co-registered anatomical images within the PET image reconstruction through a novel anatomy-guided joint prior. The characteristic of the utilized hyperbolic potential function is determined by the voxel intensity differences within the anatomical image, while the penalization is computed based on voxel intensity differences in reconstructed PET images. This formulation provides a natural solution to apply less “force” to smooth PET inter-voxel differences when having larger MR inter-voxel differences, and vice versa. Using simulated 18FDG PET scan data (simulated for the geometry of the HRRT scanner) as well as T1 MR images (9% noise applied relative to the brightest tissue), we optimized the performance of the proposed MAP reconstruction with the joint prior(JP-MAP), and compared its performance with 3D MLEM reconstruction (MLEM) and MAP reconstruction using the Bowsher prior (BP-MAP). Furthermore, we designed four hyperactive lesions (e.g. gliomas) with no corresponding anatomic counter parts in MR images for both cortical and subcortical regions in white matter, and compared lesion contrast versus noise trade off performance with BP-MAP reconstruction methods.

Results The proposed JP-MAP reconstruction algorithm resulted in quantitatively enhanced reconstructed images, as demonstrated in extensive 18FDG PET simulation studies. In particular, the proposed JP-MAP exhibited 75% and 12% reduction in noise levels (matched contrast), compared to conventional MLEM and BP-MAP techniques.

Conclusions The proposed JP-MAP PET reconstruction framework resulted in enhanced quantitative performance compared to conventional MLEM and Bowsher prior MAP techniques.

Research Support This work was supported by the 973 Program of China under grant no. 2010CB732504 and NIH grant 1S10RR023623,