US 12,254,539 B2
Systems and methods of guided PET reconstruction with adaptive prior strength
Jorge Cabello, Lenoir City, TN (US)
Assigned to Siemens Medical Solutions USA, Inc., Malvern, PA (US)
Filed by Siemens Medical Solutions USA, Inc., Malvern, PA (US)
Filed on Sep. 8, 2022, as Appl. No. 17/930,565.
Prior Publication US 2024/0095979 A1, Mar. 21, 2024
Int. Cl. G06T 11/00 (2006.01); A61B 6/00 (2024.01); A61B 6/03 (2006.01)
CPC G06T 11/008 (2013.01) [A61B 6/037 (2013.01); A61B 6/4417 (2013.01); G06T 2210/41 (2013.01); G06T 2211/424 (2013.01)] 11 Claims
OG exemplary drawing
 
1. A system, comprising:
a positron emission tomography (PET) imaging modality configured to acquire a PET dataset;
a magnetic resonance imaging (MRI) modality configured to acquire an MRI dataset; and
a processor configured to:
receive the PET dataset and the MRI dataset;
generate an MRI reconstructed image from the MRI dataset, wherein the MRI reconstructed image is registered to the PET dataset;
apply an iterative reconstruction process to the PET dataset and the MRI reconstructed image, wherein the iterative reconstruction process includes one or more similarity coefficients;
calculate an adaptive hyperparameter for each iteration of the iterative reconstruction process; and
output a reconstructed image from the iterative reconstruction process, wherein iterative reconstruction process comprises an optimization algorithm and a potential function,
wherein: (1) the optimization algorithm comprises a one-step-late (OSL) algorithm defined as:

OG Complex Work Unit Math
 where i is a line-of-response (LOR) index, M is a number of LORs, ri is scatter and random coincidences, ni is a normalization, αi is attenuation factors, gij is a system matrix, β is the adaptive hyperparameter, u is a measured activity distribution, R(u) is the potential function, and i is a voxel index; or
(2) the optimization algorithm comprises a preconditioned gradient ascent (PGA) algorithm defined as:

OG Complex Work Unit Math
 where i is a line-of-response (LOR) index, M is a number of LORs, ri is scatter and random coincidences, ni is a normalization, αi is attenuation factors, gij is a system matrix, β is the adaptive hyperparameter, u is a measured activity distribution, R(u) is the potential function, and j is a voxel index; or
(3) the optimization algorithm comprises a penalized likelihood based on a separable surrogate (PLSS) algorithm defined as:

OG Complex Work Unit Math
 where β is the adaptive hyperparameter, u is a measured activity distribution, and uj,EMn+1 is the expectation-maximization estimate of ujn smoothed according to:

OG Complex Work Unit Math
Where b is the index for voxels in a neighborhood Nj and ω is a similarity coefficient.