0685

High Resolution Perfusion Imaging using Golden Angle Radial Arterial Spin Labelling
Thomas W Okell1 and Mark Chiew1

1WIN, FMRIB, Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, United Kingdom

Synopsis

Arterial spin labeling perfusion imaging at clinical field strengths is generally confined to relatively coarse voxels (~4 mm), preventing the investigation of perfusion variations on small spatial scales and leading to problems with partial volume effects. In this work we demonstrate the ability of a golden angle radial approach combined with a regularized reconstruction technique to produce time-resolved perfusion images with isotropic voxel sizes lower than 2 mm. This was shown to improve grey matter definition and reduce partial volume effects.

Introduction

Arterial spin labeling (ASL) has become a widely used technique for non-invasive perfusion imaging1. However, due to its relatively low SNR, most brain imaging protocols at clinical field strengths (3T and below) use relatively large voxels (~4mm)1. This can be further exacerbated by through-slice blurring when using 3D readouts with long readout durations2. This not only prevents the investigation of perfusion variations on small spatial scales, but also leads to significant problems arising from partial volume effects, since the grey matter thickness is similar to the voxel size.2,3 In this work, we demonstrate the ability to obtain high isotropic spatial resolution perfusion images with ASL by combining a golden angle readout approach with a regularized iterative reconstruction.

Methods

A schematic of the pulse sequence used for high resolution perfusion imaging is given in Figure 1. A pre-saturation module is followed by a pseudo-continuous ASL (PCASL) pulse train and a spoiled 3D gradient echo golden angle radial readout. A series of images can be constructed at different postlabeling delays (PLDs) by combining sets of radial spokes within the readout period across multiple ASL preparations. This approach has been previously introduced for combined angiography and perfusion using radial imaging and ASL (CAPRIA)4,5. However, here we utilize the variable density of this radial trajectory to allow the reconstruction of perfusion images at a range of spatial resolutions: setting the maximum spatial frequency used in the reconstruction, kmax, to a low value results in low spatial resolution images with a low undersampling factor. Increasing the kmax results in higher spatial resolution images with a higher undersampling factor.

To mitigate the increased signal aliasing and noise amplification which will occur at higher spatial resolutions, two regularization approaches were tested: 1) an L1 penalty was applied to the ASL difference images to encourage sparsity in the spatial-temporal frequency (xf) domain; 2) an L1 penalty was applied in the space-time (xt) domain, with an additional L2 penalty on the temporal finite difference to encourage a smooth temporal signal evolution (referred to as xt-L2). Both approaches incorporated coil sensitivity information estimated using the adaptive combine algorithm6, and were implemented using FISTA7 with empirically determined regularization weighting factors. Results were compared to a coil-only reconstruction (iterative SENSE)8.

Three healthy volunteers were scanned under an agreed technical development protocol on a 3T Siemens Verio scanner using a 32-channel head coil. 4D CAPRIA data5 were acquired in 10min (1.1mm nominal isotropic voxels, TR/TE 9/3.4ms, variable flip angle9 2-9°, bandwidth 99Hz/Pixel, readout partial Fourier 79%). Perfusion images were reconstructed at a range of spatial resolutions with 323 ms temporal resolution. T1-weighted structural images were acquired for reference.

Results

Figure 2 compares perfusion images produced using the three reconstruction approaches. Considerable noise amplification is apparent in the SENSE-only reconstruction. Both regularization methods result in a considerable reduction in noise and signal aliasing. The xt-L2 approach produced the most robust results and was used for the remainder of this work.

Example images at each PLD are shown in Figure 3, demonstrating the expected pattern of perfusion signal. No obvious corruption of signal time-courses was observed due to the use of the L2 regularization.

Perfusion images reconstructed at various spatial resolutions are shown in Figure 4. As the spatial resolution is increased, better definition of the highly perfused grey matter can be observed, which matches closely with the structural data. At 1.4 mm isotropic voxel size, image quality begins to degrade due to noise amplification, so this data is excluded from further analysis.

The median perfusion signal (averaged over PLDs greater than one second) in white matter relative to that in grey matter is shown in Figure 5. Due to reduced partial volume effects, the average white matter ASL signal is reduced at higher spatial resolution, closer to that expected from the literature10.

Discussion and Conclusions

We have demonstrated that the combination of a golden angle radial readout with a regularized reconstruction approach allows the generation of time-resolved ASL perfusion images with isotropic spatial resolution lower than 2mm. This led to better grey matter definition and reduced partial volume effects. Although the scan time was relatively long (10 minutes), angiographic images could also be reconstructed from this same raw data5,11, increasing efficiency. The ability to acquire similar data sets in a shorter scan time will be explored in future work.

In this work, we have implicitly assumed that the perfusion signal is spatially sparse. Whilst this is true to some degree, further exploration of a wider range of regularization terms would be beneficial, along with validation against conventional approaches.

Acknowledgements

We are grateful for the facilities provided by the Oxford Acute Vascular Imaging Centre, and for funding support from the Royal Academy of Engineering. The Wellcome Centre for Integrative Neuroimaging is supported by core funding from the Wellcome Trust (203139/Z/16/Z).

References

1 Alsop DC, Detre JA, Golay X, Günther M, Hendrikse J, Hernandez-Garcia L, Lu H, Macintosh BJ, Parkes LM, Smits M, Van Osch MJP, Wang DJJ, Wong EC, Zaharchuk G. 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. Magn Reson Med 2015; 73: 102–116.

2 Chappell MA, Groves AR, MacIntosh BJ, Donahue MJ, Jezzard P, Woolrich MW. Partial volume correction of multiple inversion time arterial spin labeling MRI data. Magn Reson Med 2011; 65: 1173–1183.

3 Asllani I, Borogovac A, Brown TR. Regression algorithm correcting for partial volume effects in arterial spin labeling MRI. Magn Reson Med 2008; 60: 1362–1371.

4 Okell TW. Combined angiography and perfusion using radial imaging and arterial spin labeling. Magn Reson Med 2018. doi:10.1002/mrm.27366.

5 Okell TW. 4D Combined Angiography and Perfusion using Radial Imaging and Arterial Spin Labeling. In: Proceedings 25th Scientific Meeting, ISMRM. Hawaii, USA, 2017, p 675.

6 Walsh DO, Gmitro AF, Marcellin MW. Adaptive reconstruction of phased array MR imagery. Magn Reson Med 2000; 43: 682–90.

7 Beck A, Teboulle M. A Fast Iterative Shrinkage-Thresholding Algorithm for Linear Inverse Problems. SIAM J Imaging Sci 2009; 2: 183–202.

8 Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: Sensitivity encoding for fast MRI. Magn Reson Med 1999; 42: 952–962.

9 Schmitt P, Speier P, Bi X, Weale P, Mueller E. Non-contrast-enhanced 4D intracranial MR angiography: Optimizations using a variable flip angle approach. In: Proceedings 18th Scientific Meeting, ISMRM. Stockholm, Sweden, 2010, p 402.

10 van Gelderen P, de Zwart JA, Duyn JH. Pittfalls of MRI measurement of white matter perfusion based on arterial spin labeling. Magn Reson Med 2008; 59: 788–795.

11 Chiew M, Okell TW. Improved Golden Ratio Radial Arterial Spin Labeling Angiography Reconstruction using k-t Sparsity Constraints. In: Proceedings 26th Scientific Meeting, ISMRM. Paris, France, 2018.

Figures

Figure 1: Schematic of the CAPRIA pulse sequence used for high resolution perfusion imaging: Pre-saturation is used for background suppression, followed by the PCASL pulse train and a continuous golden angle 3D radial spoiled gradient echo readout. Using this approach, a set of images can be obtained at various different delays after labeling. In addition, with this radial trajectory images can be reconstructed at low spatial resolution with a low undersampling factor, or higher spatial resolution with a higher undersampling factor, by varying the maximum spatial frequency used in the reconstruction, kmax.

Figure 2: Comparison of perfusion images averaged over PLDs greater than 1 s reconstructed at 2.3 mm isotropic resolution using different reconstruction approaches in transverse (TRA), coronal (COR) and sagittal (SAG) planes. Extreme noise amplification is apparent in the SENSE-only reconstruction due to the high acceleration factor used (R = 3.2). Both regularized reconstructions greatly reduce noise and signal aliasing, with the xt-L2 approach resulting in the most effective noise suppression.

Figure 3: Temporal information obtained with the xt-L2 reconstruction: a) Example 2.3 mm isotropic perfusion images at each PLD for a single transverse slice; b) timecourses in the two highlighted voxels with early and late blood arrival; c) a map of the PLD at which the signal peaks in each voxel, giving coarse timing information. The expected pattern of blood arrival in larger vessels before perfusing the tissue is seen, with delayed blood arrival in the posterior circulation and watershed regions, demonstrating that the reconstruction is not overregularized in the temporal domain.

Figure 4: Comparison of perfusion images (averaged over PLDs greater than 1 s) reconstructed at different spatial resolutions. The co-registered structural image is also shown for comparison. Note the improved visualization of perfusion where the grey matter folds sharply as the spatial resolution increases (see zoomed section). At very high spatial resolution (1.4 mm isotropic), noise amplification begins to obscure image detail.

Figure 5: The median white matter ASL signal at PLDs greater than 1 s relative to that in grey matter. Data are displayed as the mean and standard deviation across subjects. As voxel size decreases, the reduced partial volume effects result in a smaller relative white matter signal, around 30% of that in grey matter. *p < 0.01 using a paired t-test.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)
0685