Impact of spatio-temporal resolution on MR-based cardiac motion correction PET-MR
Camila Munoz1, Christoph Kolbitsch1,2, and Claudia Prieto1

1Department of Biomedical Engineering, King's College London, London, United Kingdom, 2Division of Medical Physics and Metrological Information Technologies, Physikalisch-Technische Bundesanstalt, Berlin, Germany

Synopsis

MR-based PET motion correction has been shown to improve image quality in cardiac PET-MR imaging. Here we present a numerical simulation study analysing the impact of temporal and spatial resolution of motion fields on the final image quality of myocardial perfusion PET scans in order to find the most efficient parameters yielding accurate cardiac motion compensation in the shortest possible scan time. Results show that cardiac motion correction is important for accurate assessment of myocardial lesions and that temporal resolution of the motion fields can be strongly optimised without losing diagnostic accuracy, reducing the total exam time in PET-MR imaging.

Introduction

Simultaneous PET-MR offers the possibility to use high-resolution MR images for motion correction of PET data. Several studies have shown that this approach leads to a strong improvement in PET image quality. Here we present a numerical simulation study analysing the impact of temporal and spatial resolution of MR motion fields on the final image quality of myocardial perfusion PET scans in order to find the most efficient motion field parameters yielding accurate cardiac motion compensation in the shortest possible scan time.

Methods

A numerical representation of a human torso and parameterised models of beating heart motion are obtained using a commercially available software (XCAT(1)). XCAT generates emission maps, where each voxel represents a tissue dependant standardized uptake value (SUV), and attenuation maps for each motion state, that are then used as input for PET simulations. Reconstructed PET images were assessed based on metrics used for clinical diagnosis including detectability of damaged segments in standard polar plots computed from the left ventricular myocardium(2), and extent and degree of transmurality (i.e. fraction of the wall affected) of the lesion.

Four myocardial lesions with different transmurality (100%, 50% and 25%), extension and location (anterior and lateral wall) were simulated using realistic 18F-FDG uptake values(3) in XCAT (Fig.1). Cardiac motion was simulated based on a normal cardiac cycle with a duration of 1s. Analytical PET simulations were performed in STIR(4) using the geometry of the Siemens Biograph mMR scanner (60cm bore size, 8 rings of LSO crystal detectors, 258mm axial FOV). Motion fields of five temporal (20, 10, 5, 2 and 2* cardiac phases) and four spatial (2, 4, 5.68 and 9.84mm3 isotropic voxel size) resolutions were studied. The 2* non-uniform division of the cardiac cycle groups two thirds of the cardiac cycle around maximum relaxation in one cardiac phase and the remaining third as the second phase. Image reconstruction was performed using a motion compensated algorithm(5) (21 subsets, 3 iterations) with a voxel size of 2.03×2.08×2.08mm3, a matrix size of 127×285×285 and a 4mm isotropic Gaussian post-filtering. Additionally, a motion-free image was simulated and reconstructed for reference purposes, and an uncorrected reconstruction was performed for further comparison.

Results

For both transmural and non-transmural perfusion defects, results showed that although image quality degrades when reducing temporal and spatial resolution of the motion fields, non-viable myocardium can be detected even without motion correction (Fig.1). For defects 1 (transmural) and 2 (50% transmural), a resolution of 10mm and 2* phases allowed to detect more of 90% of the non-viable segments. For defects 3 (25% transmural) and 4 (50% transmural), a resolution of 2mm and 2* phases is required to detect more than 90% of the non-viable segments correctly. For the transmurality analysis both (2mm, 20 phases) and (2mm, 2* phases) lead to accurate results; considering defects 2, 3 and 4, an average error of 7.17 ± 3.44% and 5.58 ± 1.53% compared to the motion-free reference was obtained respectively. For defect 4 (50% transmural), located in an area of large cardiac motion amplitude, when 2 phases are used or no motion correction is performed, the lesion is erroneously detected as transmural (Fig. 3)

Conclusion

We have presented a simulation study to assess the impact of spatial and temporal resolution of motion fields for MR-based beating heart motion corrected cardiac perfusion PET imaging. Results show that non-viable myocardium can be detected even without motion correction however motion correction is essential for accurate estimation of the size and degree of transmurality of perfusion defects. The non-uniform two-phase division of the cardiac cycle allowed for achieving less than 10% error in the estimation of extent and transmurality of all simulated defects using a spatial resolution of 2mm. Thus, temporal resolution of the motion fields can be significantly reduced without losing diagnostic accuracy. The subsequent reduction of time allocated for MR-based motion estimation would potentially allow for acquiring diagnostic MR information simultaneously with PET, reducing total exam time in PET-MR clinical routine.

Acknowledgements

This work was supported by the EPSRC Centre for Doctoral Training in Medical Imaging.

References

1. Segars et al. 4D XCAT phantom for multimodality imaging research. Med Phys 2010, 37(9):4902-4915.

2. Cerqueira et al. Standardized Myocardial Segmentation and Nomenclature for Tomographic Imaging of the Heart. JCMR 2002, 4(2):203-210.

3. Wang et al. Standardized uptake value atlas: Characterization of physiological 2-Deoxy-2-[18F]fluoro-d-glucose uptake in normal tissues. Mol Imaging Biol 2007, 9(2):83–90.

4. Thielemans et al. STIR: software for tomographic image reconstruction release 2. Phys Med Biol 2012, 57(4):867-883.

5. Tsoumpas et al. The effect of regularization in motion compensated PET image reconstruction: a realistic numerical 4D simulation study. Phys Med Biol 2013, 58(6):1759–73

Figures

Fig1.Simulated defects (first column) and corresponding example polar plots (columns 2 to 5, damaged segments are indicated with a black X). When using 2* phases and an appropriate spatial resolution, more of 90% of the non-viable segments can be detected. (A: anterior, S: septal, L: lateral, I: inferior)

Fig2.Transmurality estimation for 50% transmural defects located in areas of small (defect 2) and large (defect 4) cardiac motion amplitude. For all defects resolution can be reduced to 2*phases-2mm without losing accuracy. For defect 4, the lesion appears erroneously as 100% transmural in the uncorrected and 2-phases motion corrected reconstructions.

Fig3.Short axis slices for motion free, uncorrected and two motion compensated reconstructions for defect 4 (columns 1, 2) and profiles through the lesion (column 3). For the uncorrected reconstruction, the remaining healthy myocardium cannot be identified so the lesion appears erroneously as transmural.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
2195