Introduction

Restauration of blood flow to an ischemic organ is essential to prevent irreversible tissue injury. To this end, reliable estimation of cerebral blood flow is crucial for a precise diagnosis and patient therapy. Recent introduction of PET-MRI hybrid technology allows the simultaneous acquisition of PET and MRI data. Multi-parametric PET-MRI can be used to quantitatively evaluate physiological parameters, and to identify different areas within the ischemic territory. PET with [¹⁵O]H₂O is the reference method to quantitatively assess cerebral perfusion. MRI cerebral perfusion can also be evaluated using DSC-MRI (dynamic susceptibility contrast MRI) with an injection of a paramagnetic contrast agent.

Methods

Longitudinal cerebral perfusion assessment using a hybrid PET-MRI scanner Biograph mMR (Siemens Healthcare, Erlangen, Germany) was performed in a minimally invasive endovascular non-human primate (NHP – Macaca fascicularis) model of stroke under continuous veterinarian monitoring. The study was approved by the ethic committee of the institution (APAFIS#8901-2016032116237108) and strictly followed the European guidelines for animal experiment. In this model, a transient occlusion of the middle cerebral artery (MCA) was performed. Imaging data were acquired before ischemia (“baseline” imaging session) and after recanalization of the ischemic area (“post-reperfusion” imaging session). In this pilot phase, two animals were scanned at baseline session (NHP#1 and NHP#2) and one animal at post reperfusion session (NHP#1). PET data were corrected from attenuation using CT images. Kinetic modeling of the [¹⁵O]H₂O PET was performed to compute CBF parametric maps. An image derived input function (IDIF) using a region of interest (ROI) placed in the aortic arch was used for the kinetic modeling. DSC-MRI data were processed using the software Olea Sphere® (Olea Medical, La Ciotat, France) and 4 post-processing algorithms were compared. DSC-MRI requires the measurement of the arterial input function (AIF) and the deconvolution of the tissue concentration time curve. One of the most accepted deconvolution methods is the use of singular value decomposition (SVD) and three variants of SVD were used (sSVD, cSVD, oSVD). A recently developed method, based on a probabilistic approach, the Bayesian algorithm were also used to compute MRI based parametric maps of blood flow, MRI-CBF. An automatic (based on clustering of arterial voxels) or manual selection of the AIF was also used to derive the MRI-CBF maps. Brain regions of interests (ROIs) consisting in 6-mm circles were manually placed over the cortex, basal ganglia, and white matter areas in both the affected and unaffected hemispheres, using the T1-weighted scans (72 ROIs for NHP#1 and 83 ROIs for NHP#2). For each ROI, PET- and MRI-CBF values were obtained. The regional CBF values MRI-CBF obtained with the DSC-MRI and the four post-processing algorithms, were compared by linear correlation with the PET-CBF, and Bland-Altman plots to characterize similarities and differences between methods. NHP#1 and NHP#2 ROIs data were pooled together. To reduce intersubject variability, PET- and MRI-CBF data were normalized, each ROI data was expressed as percentage of the mean of all ROIs for each NHP.

Results

NHP#1 lesions observed on the diffusion-weighted imaging (DWI) at the post-reperfusion imaging session, and corresponding PET- and MRI-CBF maps are represented on Figure 1. Coefficient of correlation R² obtained in the regression plots are represented in Table 1. The two MRI post processing methods that give the best correlation with PET-CBF data are outlined in red. A good correlation between MRI methods is obtained. By comparison to the reference methods PET-CBF, automatic and manual AIF with the Bayesian algorithm showed the best correlation. The plots of these two regressions shown in Figure 2. Presence of low and high values of CBF at the post reperfusion imaging session improves linear link between PET and MRI-CBF values, and reduce dispersion. The slopes of the correlations show either an overestimation or underestimation of the MRI-CBF against the reference method PET-CBF. These observations are confirmed by the Bland-Altman plots shown in Figure 3. In this case, automatic selection of the AIF gives higher bias in both baseline and post reperfusion imaging sessions, compared to manual selection of the AIF that gives a bias inferior to 10%.

Discussion

Correlation with the automatic AIF and Bayesian method versus PET-CBF is close to the one obtained with manual AIF (R²=0,50 versus R²=0,47) but when looking at absolute values, manual AIF gives the closest estimation of MRI-CBF compared to PET-CBF. A larger sample is needed to confirm the value of the Bayesian algorithm for longitudinal studies of CBF and CBV.

Acknowledgements

The authors would like to thank Siemens Healthcare for providing the prototype sequence used in this work.

References

No reference found.