Artifact-free quantitative cardiovascular PET/MR imaging: An impossible dream?

Abstract

Building a winning team requires a bit of complementarity and convincing common goals to overcome the frustration of being tied together. In the journey of cardiac PET/MRI, this frustration has a name: less optimal attenuation correction (AC) as compared to its PET/CT counterpart where CT-based AC works reasonably well. AC has been a key component in the development of nuclear cardiology providing at the same time more reliable assessment of regions with relative tracer abnormalities and the ground for accurate quantification of tracer uptake.1 It is therefore no wonder that reproducing the same level of confidence in AC using MRI as achieved using CT has been a major drive in the research over PET/MRI in general and in cardiovascular imaging applications, in particular.MRI segmentation-based approaches have focused on deriving attenuation maps from basic MR sequences, such as T1-weighted or DIXON sequences; the latter consists of in-and opposed-phase images enabling segmentation in four tissue types (air, soft tissue, lung, and fat).2 Among the limitations of this approach, when using these sequences, is the inconsistent differentiation between air and bone densities. This is important when having in mind the position of the patient during thoracic PET/MRI with the arms laying aside the body to promote optimal positioning of the radiofrequency surface coils and the relatively small field of view of MR images (only 45-50 cm compared to 60-70 cm on contemporary PET/CT systems) that systematically brings about truncation artifacts. A number of strategies enabling to incorporate bone density have been proposed and potential solutions, such as ultrashort echo time and zero echo time sequences, have been proposed and refined and are finding their way to the clinic in the context of brain imaging. 3,4 In theory, incorporating an increased number of tissue types in the MRI-derived attenuation map would allow reducing quantification errors within the range of 5% compared to CT-derived attenuation maps as proposed by Keereman et al using five tissue types. 5The recent advances in detector technology embedded in the latest designs of PET/MRI systems have contributed to the revival of an old concept referred to as joint estimation of activity and attenuation maps from emission data. This method has capitalized on time-of-flight capability to refine the positioning information on PET lines of responses for improved signalto-noise ratio. 6 The attenuation maps could therefore be derived from the emission data using maximum likelihood reconstruction of activity and attenuation (MLAA) type of algorithms that are penalized using priors to reduce the cross-talk between activity and attenuation estimations. The scaling issue inherent in this approach has been recently dealt with using MLAA with a Gaussian mixture model approach.7 Noteworthily, this category of techniques is now routinely integrated by one of the manufacturers for the compensation of truncation artifacts.

Hybrid PET- and MR-driven attenuation correction for enhanced 18F-NaF and 18F-FDG quantification in cardiovascular PET/MR imaging

“…Simultaneous PET and Magnetic Resonance (MR) systems have recently enabled the co-registration of PET images of highly specific molecular activity with multi-parametric MR images of superior soft-tissue resolution (7)(8)(9)). Yet MR-based AC (MR-AC) may not be as accurate as CT-AC, due to MR's inherent limitations in ACFs quantification, especially for highly attenuating tissues, such as cortical bones (10)(11)(12). Currently, PET/MR scanners employ MR-Dixon sequences to segment all tissues into four classes, namely air, lungs, fat and soft-tissues, each assigned a discrete ACF value (13).…”

“…Currently, PET/MR scanners employ MR-Dixon sequences to segment all tissues into four classes, namely air, lungs, fat and soft-tissues, each assigned a discrete ACF value (13). Thus, the bones are notably neglected in 4-class MR-AC leading to PET quantification errors in adjacent-to-bone regions (12).…”