Impact of calibration method on the reproducibility of CBF mapping using multiple post-labeling-delay PASL
Joana Pinto1, Pedro Vilela2, Michael A. Chappell3, and Patrícia Figueiredo1

1ISR-Lisboa/LARSyS and Department of Bioengineering, Instituto Superior Técnico – Universidade de Lisboa, Lisbon, Portugal, 2Imaging Department, Hospital da Luz, Lisbon, Portugal, 3Institute of Biomedical Engineering, University of Oxford, Oxford, United Kingdom

Synopsis

Absolute CBF quantification using ASL requires the normalization of the control-label difference images by the equilibrium magnetization, M0. A voxelwise calibration method is currently recommended for single post-labelling-delay (PLD) PCASL. However, the impact of using an M0t map obtained directly from the ASL data, with no need for an extra scan, by fitting a saturation-recovery curve to the control image time-series in multiple-PLD PASL remains to be investigated. Here, we show that, using this type of acquisition, voxelwise calibration significantly reduced inter- and intra-subject variability in gray matter CBF measurements relative to methods based on a reference tissue.

Purpose

In order to obtain cerebral blood flow (CBF) measurements in absolute units using ASL, control vs label difference images are normalized by the equilibrium magnetization of arterial blood (M0a)1,2. This is usually estimated from a calibration image that is separately acquired, by extrapolating the value of M0a from the equilibrium magnetization measured in a tissue (M0t). In pulsed ASL (PASL) acquisitions at multiple post-labeling-delays (PLD) with PICORE labeling, the control images follow a saturation-recovery curve, allowing the estimation of M0t and T1 maps without the need of an extra scan. M0a can then be computed using different methods, either based on a reference tissue, particularly cerebrospinal fluid (CSF) or white matter (WM), or on the voxelwise M0t values. Although different calibration methods have been previously compared1,2, their impact on the reproducibility of CBF measurements has not yet been reported. Moreover, no systematic comparison of the results obtained using control saturation-recovery data in multiple-PLD PASL has been performed. Here, we investigate the impact of different calibration methods on multiple-PLD PASL on CBF quantification and reproducibility.

Methods

Nine healthy volunteers were studied on a 3T Siemens Verio system equipped with a 12-channel-receive head RF coil in two sessions. ASL images with 9 contiguous axial slices with 3.5x3.5x7.0mm3 resolution were obtained using PICORE-PASL with TR/TE=2500ms/19ms, Q2TIPS saturation limiting the labeling bolus width to 750ms, 11 PLD values between 400 and 2400ms, in steps of 200ms, and 8 label/control repetitions for each PLD. Two PASL datasets were acquired with/without macroflow crushing (crushed/non-crushed)3. A 1mm isotropic resolution T1-weighted structural image was collected using MPRAGE. ASL data pre-processing included: motion and slice-time correction; control magnetization averaging at each PLD (control time series); and control-label magnetization subtraction and averaging at each PLD (difference time series). M0t and T1 maps were obtained by fitting a saturation-recovery curve to the control time series. T1 maps were used for co-registration with the structural image, which was segmented into gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF). An extended kinetic model with/without intravascular arterial compartment was fitted to the non-crushed/crushed difference time series using BASIL4,5, with T1a=1.6s, T1t=1.3s, τ=750ms, and α=0.9. Calibration of CBF estimates was then performed using the following three methods:

Method 1) Reference Tissue:

Method 1.1) CSF: $$$M_{0a}=(<M_0>_{CSF}\times \exp(TE\div T^*_{2CSF}-TE\div T^*_{2a}))\div λ_{CSF}$$$, where T*2a=50ms, T*2CSF=400ms and λCSF=1.15

Method 1.2) WM: $$$M_{0a}=(<M_0>_{WM}\times \exp(TE\div T^*_{2WM}-TE\div T^*_{2a}))\div λ_{WM}$$$, where T*2a=50ms, T*2WM=50ms and λWM =0.82

Method 2) Voxelwise: $$$M_{0aWM}(i)=M_{0WM}(i)\div<λ>_{WM}$$$, with <λ>WM=0.82 and $$$M_{0aGM}(i)=M_{0GM}(i)\div<λ>_{GM}$$$, with <λ>GM= 0.98

The median of CBF across GM and WM, and their GM-to-WM ratio, were computed in each dataset. Repeated measures ANOVA was performed across subjects to test for the effects of calibration method and data type. Reproducibility of GM CBF measurements was assessed by computing the inter- and intra-subject coefficients of variation (CVinter and CVintra) and the two-way mixed intraclass correlation coefficient (ICC)3. A GM region was defined by intersection between the subject’s GM mask and 9 MNI GM regions of interest (ROIs). Repeated measures ANOVA was performed across the GM ROIs to test for the effects of calibration method and data type.

Results

Representative CBF maps obtained using the three calibration methods are shown in Fig.1. The subject and group averages of CBF in GM and WM, as well as the GM-to-WM CBF ratio, are displayed in Fig.2. Significant differences were observed in GM CBF between the voxelwise and the reference tissue methods, with voxelwise calibration producing a greater ratio than the reference tissue methods. The reproducibility results for the GM CBF measures are presented in Fig.3. All CV and ICC values are within the intervals of good/acceptable reproducibility (CVintra<33% and ICC>0.4)4. A main effect of method was found for CVinter and CVintra, with the voxelwise method in general producing the lowest variability. No effects were found for ICC, which can be explained by the interplay between the reductions in both inter- and intra-subject variability.

Conclusion

We found that voxelwise calibration produced significantly reduced inter- and intra-subject variability in GM CBF measurements relative to methods based on a reference tissue, when using M0t maps estimated from the control saturation-recovery curve in multiple-PLD PICORE-PASL. Besides the reduced variability, the voxelwise method is also advantageous because it takes into account RF inhomogeneity and differences in transverse relaxation, and it does not depend on segmentation to obtain a reference tissue mask. Although the ASL white paper recommends voxelwise calibration for single-PLD pseudo-continuous ASL6, our results further indicate that a voxelwise approach is also preferable in multiple-PLD PASL when using the saturation-recovery curve of the control images for M0t estimation.

Acknowledgements

This work was funded by FCT grants PTDC/BBB-IMG/2137/2012 and Pest-OE/EEI/LA0009/2013, Hospital da Luz SA and European Union COST Action BM1103

References

1. M. Cavusoglu et al. “Comparison of pulsed arterial spin labeling encoding schemes and absolute perfusion quantification”, 27, Magnetic Resonance Imaging, 2009.

2. Y. Chen et al., “Impact of equilibrium magnetization of blood on ASL quantification”, Proceedings ISMRM 2011.

3. I. Sousa et al., “Reproducibility of the quantification of arterial and tissue contributions in multiple postlabeling delay arterial spin labeling”, 40 (6), Journal of Magnetic Resonance Imaging, 2014

4. M. A. Chappell et al., “Separation of Macrovascular Signal in Multi-inversion Time Arterial Spin Labelling”, 63, Magnetic Resonance in Medicine, 2010.

5. M. A. Chappell et al., “Variational Bayesian inference for a non-linear forward model”, 57 (1), IEEE Transactions on Signal Processing, 2009.

6. D. C. Alsop et al., “Recommended Implementation of Arterial Spin-Labeled Perfusion MRI for Clinical Applications: A Consensus of the ISMRM Perfusion Study Group and the European Consortium for ASL in Dementia”, 73, Magnetic Resonance in Medicine, 2015

Figures

Fig 1. CBF maps obtained using the three calibration methods tested, for both non-crushed and crushed data, for one illustrative slice of one subject.

Fig 2. Median CBF in GM and WM, and corresponding GM-to-WM ratio, obtained for each subject and session (S1 and S2), and respective group mean, using the three calibration methods, and the two data acquisition types (error bars represent SD, p<0.05).

Fig 3. Reproducibility of GM CBF measurements, using the three calibration methods and the two data acquisition types (error bars represent SD, p<0.05).



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0424