Target Audience

Physicians and Physicists who are interested in performing PET/MR neuroimaging study

Objective

Provide an overview of current status of MR based attenuation correction for PET imaging

Introduction

In PET imaging, an annihilation of an emitted positron and with an electron produces two 511 KeV photons moving inthat move in opposite directions. These photons travel through the tissue before reaching PET detectors. The absorption and scatter caused by the photon-tissue interaction leads to photon attenuation. Photon attenuation depends on the spatially varying both the electron density and tissue thicknesshe thickness of the tissue and these effects vary spatially. Photon attenuation can result in as high as 90% signal reduction. Therefore, attenuation is by far the largest correction required for quantitative PET imaging. PET attenuation correction methods require the knowledge of the spatial distribution of tissue attenuation coefficients within the PET field of view. MR imaging does not provide direct information on electron density needed by PET attenuation correction. While PET/MRI is FDA approved for clinical use, MRI-based attenuation correction (MRAC) methods have not been well accepted for clinical trials. In the past several years, numerous approaches have been proposed to develop attenuation correction for PET/MR imaging. These methods can be roughly classified into two major categories: atlas based and direct imaging based approaches. The atlas-based methods usually derives a computational relationship from a group of observed CT and MRI image pair using population data, which can be generalized for future deployment when only MRI is available. Direct MR imaging method use Dixon, ultra-short echo (UTE) or zero echo time (ZTE) MR scans to derive PET attenuation correction maps without or with little population based information. In this category of methods, individual patient MR images are segmented into several tissue classes, followed by assignment of a constant linear attenuation coefficients (LAC) toIn the early effort each tissue class or deriving conversion factors to convert MR signal/relaxation rates to CT HU for continuous LAC in bone. Since tissue segmentation has been widely used in the direct imaging approach, this category of methods are sometimes referred as segmentation based approach in some literatures.

Atlas based approaches

Depending on how pseudo-CT images were generated, the atlas based methods can be roughly classified into voxel based, patch based, and machine learning based pseudo-CT generation subgroups in this review. The basic premises of atlas based AC approaches are 1) each individual patient’s anatomy can be well represented by the population data; 2) morphological similarity of patient’s MR images to the atlas images can result in CT HU similarity. Data with either abnormal anatomy and/or unusual tissue density (e.g. bone density) that is very different from population average may lead to large AC errors. Moreover, atlas based methods usually involves complex computation which may be time consuming. Thus far, atlas based methods have been well tested in neuro-imaging. In atlas based methods, continuous LACs are available and the population averaged information provide robustness to imaging artifacts and noise. However, atlas based approaches cannot account for inter-subject variations. Since age, gender and race can impact bone thickness and density, a single atlas based CT estimation may not be adequate for all patients. Separate MR-CT pair database is needed for AC in children.

Direct imaging based approaches

Direct imaging methods utilize MR images acquired using Dixon, UTE or ZTE images to derive PET AC maps without atlas alignment and complex pseudo-CT generation. The proposed methods can be roughly classified into two sub-groups: segmentation-only and segmentation+MR-CT conversion. The direct imaging methods are fast and can account for patient variability. However, the direct imaging can be negatively impacted by image artifacts. Dixon images are very quick to acquire (<20 seconds). However, since it does not provide bone information, large errors are expected, especially in brain or pelvic PET imaging. It usually take several minutes to acquire UTE or ZTE images. Both of UTE and ZTE can be used for bone segmentation. However, the direct imaging with segmentation only methods under-represent the continuous electron density in PET attenuation correction. Direct imaging methods with segmentation and continuous LAC value conversion can achieve PET AC accuracy on par with the atlas based MRAC[y1] . This class of method relies on direct UTE or ZTE MR imaging for tissue segmentation and a conversion using either DUTE R2* or the inverse of logarithm of ZTE signal to CT HU for continuous LAC values in bone. Since the MR-to-CT conversion is predefined, these methods are much faster than the atlas based method. These methods can account for individual subject variations better than the atlas based methods. They can be easily translated to whole body. These methods can be directly applied to any age group, including children. Attenuation map generation is rapid. Due to imaging noise and artifacts, they are not as robust as the atlas based method. Direct imaging with segmentation only approaches have large AC errors due to discrete LAC values. The direct imaging with segmentation and MR-CT conversion address this problem and have improved PET AC accuracy.

Conclusions

MR-AC in adult brains with normal anatomy has been solved to an acceptable degree, with errors smaller than the quantification reproducibility in PET imaging. Cortical region and regions near the skull base usually demonstrate larger errors. Caution needs to be used when interpreting results in these regions. Clinical evaluation across vendors and centers are needed.

Acknowledgements

No acknowledgement found.

References

No reference found.