Impact of MR-based PET motion correction on the quantification of myocardial blood flow: an in-vivo simultaneous MR/PET study
Yoann Petibon1, Behzad Ebrahimi1, Timothy G Reese1,2, Nicolas Guehl1, Marc D Normandin1, Nathaniel M Alpert1, Georges El Fakhri1, and Jinsong Ouyang1

1Center for Advanced Medical Imaging Sciences, Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, United States, 2Athinoula A. Martinos Center, Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, United States

### Synopsis

Dynamic PET imaging enables absolute quantification of myocardial blood flow (MBF). However, motion of the heart during imaging deteriorates the accuracy of PET MBF measurements. Simultaneous MR/PET makes it possible to compensate PET images for motion by incorporating MR-based motion information inside the PET reconstruction process. In this study, we propose and assess the impact of a tagged-MRI based PET motion-correction technique for improved PET MBF quantification using an in-vivo simultaneous MR/PET study.

### Introduction

PET is the reference modality for absolute quantification of myocardial blood flow (MBF). MBF estimation is typically performed by measuring changes in the spatio-temporal bio-distribution of the radiotracer by reconstructing a dynamic series of PET images, followed by compartmental analysis using appropriate kinetic models. Yet, the accuracy of MBF measurements is substantially reduced by heart motion during the acquisition, which introduces blurring and biases in the dynamic PET images1. Gating, which is frequently used to mitigate motion blurring in ‘static’ PET, is difficult to translate to dynamic PET due to the noise increase associated with rejecting data in short frames. Simultaneous MR/PET, a novel hybrid modality with promising cardiovascular applications2, provides a solution to the motion problem in PET since simultaneously acquired MRI-based motion information can be used to compensate PET data for motion without penalizing image SNR3. In this study, we propose and assess the impact of a tagged-MRI based PET motion-correction technique for improved PET MBF quantification using an in-vivo simultaneous MR/PET study.

### Methods

MR/PET imaging: An anesthetized healthy pig underwent a dynamic study on a simultaneous MR/PET scanner (Siemens Biograph mMR). Following the acquisition of a MR-based attenuation map, list-mode PET data were collected for 20min following injection of 100MBq 18F-Flurpiridaz, a cardiac perfusion radiotracer4. To track cardiac motion, tagged-MRI data were acquired simultaneously using an ECG-triggered 1-1 SPAMM5 cardiac-gated GRE sequence (TE/TR=2.1/60ms, flip angle=5°, in-plane resolution=2.3×2.3mm2, slice thickness=8mm, tagging distance=8mm). Ten tagged-MR volumes (one per cardiac phase) were acquired, and three orthogonal SPAMM directions were imaged sequentially. Respiration-induced heart motion was neglected due to the specific anatomy of the pig’s thorax.

MR-based PET motion-correction: Motion fields from a given cardiac phase to a selected reference phase were calculated by nonrigid B-spline registration of tagged-MR volumes6. The resulting motion fields were incorporated into a modified OSEM PET reconstruction algorithm3 which iteratively estimates a motion-corrected PET image $\bf\widehat{\rho}_{MC}^{it}$ using $$\bf\widehat{\rho}_{MC}^{it+1}=\frac{\widehat{\rho}_{MC}^{it}}{\sum_tM_{ref\rightarrow{t}}^{T}G^{T}A_tN1_I}\times\sum_{t=1}^\it{M}\bf\left[M_{ref\rightarrow{t}}^{T}G^{T}\frac{y_t}{GM_{ref\rightarrow{t}}\widehat{\rho}_{MC}^{it}+(AN)^{-1}(\overline{S}+\overline{R})}\right]$$ Where $\bf{y_t}$ contains the PET data measured in phase t=[1..M], $\bf{M_{ref\rightarrow{t}}}$ is the tagged-MRI based motion-warping operator, $\bf{G}$ models the forward-projection, $\bf{A_t}$ and $\bf{N}$ respectively contain the attenuation correction coefficients for phase t and the detector normalization factors, and $\bf{\overline{S}}$ ($\bf{\overline{R}}$) are the scatters (randoms) coincidences distributions. This algorithm reconstructs a motion-free volume depicting the radiotracer’s distribution in the reference phase (end-diastole) using PET data from all phases and thus, without increasing noise.

List-mode PET data were framed into a series of 12×5s, 8×15s, 4×30s and 5×60s dynamic frames. The data in each frame were then sorted into 10 cardiac phases based on R-wave triggers inserted in the list-mode. Each frame was reconstructed as ‘Ungated’ (all data included regardless of motion), ‘Gated’ (only end-diastolic data -~40% of the cardiac cycle), and ‘Motion-Corrected’ (using the proposed method).

MBF quantitation: The kinetics of 18F-Flurpiridaz were described by a one-compartment model in which the myocardial time-activity-curve (TAC) $C_{myo}(t)$ is defined by the following differential equation $$\frac{\text{d}{C_{myo}(t)}}{\text{d}t}=K_1C_{a}(t)-k_2C_{myo}(t)$$ Where $C_{a}(t)$ is the arterial input function, commonly approximated by the left-ventricle (LV) TAC $C_{LV}(t)$ as used hereinafter, $K_1$ [mL.min-1.g-1] the uptake rate and $k_2$ the washout rate [min-1]. $C_{LV}(t)$ was calculated using a volume-of-interest (~1x1x2.5cm3) centered in the LV. Solving the differential equation yields $$C_{myo}(t)=C_{LV}(t)*{K_1e^{-k_2t}}$$ Due to finite resolution effects, the myocardial TAC measured by PET is contaminated by spill-over from the LV and the RV blood-pools $$PET(t)=f_{LV}C_{LV}(t)+f_{RV}C_{RV}(t)+(1-f_{LV}-f_{RV})C_{LV}(t)*K_1e^{-k_2t}$$ Where $f_{LV}$ ($f_{RV}$) is the fractional spill-over from LV (RV), accounting for contribution of blood in the measured TACs, and $K_1=EF\times{MBF}$ where $EF=0.94$4 is the tracer's extraction fraction. Voxel-wise MBF maps were calculated by non-linear least-square fitting of myocardial TACs using the aforementioned PET model for Ungated, Gated and Motion-Corrected dynamic data.

### Results

Fig. 1 depicts the motion fields estimated between the end-systolic and end-diastolic phases using tagged-MRI. Fig. 2 shows short-axis PET images obtained from the last 2min dynamic data along with myocardium-to-blood contrasts (Ci) calculated in 4 ROIs. Fig. 3 shows TACs and model fits obtained from the dynamic data within one ROI in the inferior wall. Finally, Fig.4 shows short-axis MBF maps estimated from the dynamic data, along with mean MBFs calculated in 4 ROIs.

### Discussion and Conclusion

The results obtained show that the proposed method improved the spatial resolution of cardiac PET images (e.g. papillary muscle, Fig. 2) and increased the reconstructed myocardium-to-blood contrast (16-37%) -particularly in the lateral wall- as compared to Ungated images without gating-associated noise increase. This improved image quality resulted in higher (5-35%) and more uniform MBF values within healthy myocardial tissue as compared to Ungated dynamic data (Fig. 4), yielding mean MBFs similar to Gated ones without increasing pixel-wise MBF variability.

### Acknowledgements

This work was supported in part by NIH grants R01-HL118261 and R01-HL110241.

### References

1. R. Klein and R. S. Beanlands, "Quantification of myocardial blood flow and flow reserve: technical aspects," Journal of nuclear cardiology 17, 555-570 (2010).

2. O. Ratib and R. Nkoulou, "Potential applications of PET/MR imaging in cardiology," Journal of Nuclear Medicine 55, 40S-46S (2014).

3. Y. Petibon, C. Huang, J. Ouyang, T. G. Reese, Q. Li, A. Syrkina, Y.-L. Chen and G. El Fakhri, "Relative role of motion and PSF compensation in whole-body oncologic PET-MR imaging," Medical physics 41, 042503 (2014).

4. S. Nekolla, S. Reder, A. Saraste, T. Higuchi, G. Dzewas, A. Preissel, M. Huisman, T. Poethko, T. Schuster and M. Yu, "Evaluation of the novel myocardial perfusion positron-emission tomography tracer 18F-BMS-747158-02 comparison to 13N-ammonia and validation with microspheres in a pig model," Circulation 119, 2333-2342 (2009).

5. L. Axel and L. Dougherty, "MR imaging of motion with spatial modulation of magnetization," Radiology 171, 841-845 (1989).

6. S. Y. Chun and J. A. Fessler, "A simple regularizer for B-spline nonrigid image registration that encourages local invertibility," Selected Topics in Signal Processing, IEEE Journal of 3, 159-169 (2009).

### Figures

Figure 1: Same transverse tagged-MRI slice taken at A) end-diastole and B) end-systole. The motion fields estimated between the 2 phases is depicted by the yellow arrows in B).

Figure 2: Short-axis static PET images obtained using A) Ungated, B) Gated and C) Motion-Corrected. Myocardium-to-blood contrasts calculated in the 4 ROIs defined in D) are shown. Yellow arrows point to a papillary muscle barely visible in A); green arrows point to the inferior wall which particularly benefits from motion-correction.

Figure 3: Ungated, Gated and Motion-Corrected time-activity curves and model fits in a ROI located in the inferior wall.

Figure 4: Short-axis MBF maps estimated from Ungated (A), Gated (B) and Motion-Corrected (C) dynamic PET data. Mean±SD MBF values calculated in the 4 ROIs defined in D) are shown.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0880