PET/MR: improvement of the UTE ?-maps using modified MLAA

For a quantitative analysis in positron emission tomography (PET) or single-photon
emission computed tomography (SPECT), attenuation correction (AC) is mandatory. CTscans
or transmission scans are common tools for determination of the attenuation ?-map,
but in the case of a PET/MR hybrid system it is difficult to associate one of these
scans. Many techniques have been developed in order to improve AC for PET/MR. Some
methods are based on template- or atlas techniques, other methods apply a segmentation
technique based on Dixon or UTE (Ultrashort Echo Time) MR to create the ?-map, followed
by a standard OSEM reconstruction (OSEM/DIXON and OSEM/UTE). A different approach
for AC has been developed by employing the emission sinogram data in the ?-map derivation.
In this context, we modified the iterative MLAA (Maximum-Likelihood reconstruction
of Attenuation and Activity) algorithm to improve the resulting emission image from
the PET/MR system. We constrained the attenuation map update using the UTE ?-map and
the T1-weighted (T1w) MR image in order to improve convergence towards a solution.
Results show that the modified MLAA algorithm improved the estimated emission image
compared to standard OSEM/UTE and OSEM/DIXON. In certain regions of the brain, in
particular close to the skull and the air cavities, the modified MLAA algorithm generated
less error than OSEM/UTE and OSEM/Dixon. The modified MLAA algorithm is able to compute
an attenuation ?-map that is slightly more similar to the aligned CT ?-map than the
UTE ?-map.