2978
Towards 1 min brain morphometry - evaluating compressed-sensing MPRAGE1Center for Brain Science, Harvard University, Cambridge, MA, United States, 2Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States, 3Department of Psychology, Harvard University, Cambridge, MA, United States, 4Advanced Clinical Imaging Technology, Siemens Healthineers, Lausanne, Switzerland, 5Department of Radiology, University Hospital, Lausanne, Switzerland, 6École Polytechnique, Fédérale de Lausanne, Lausanne, Switzerland
Synopsis
A new MR scan acceleration method, employing incoherent undersampling and compressed-sensing reconstructions, can reduce the scan time for a 1.0 mm MPRAGE to 60-90 seconds, depending on acceleration level. We have validated the morphometrics from a prototype compressed-sensing MPRAGE sequence, with different levels of acceleration, with those from conventional MPRAGE scans. Surfaces created from the compressed-sensing MPRAGE images match those from the conventional scan well. Bulk morphometric values such as total gray and white matter volume, and average cortical thickness are similar to those determined with a conventional MPRAGE scan.
Introduction
Automated MRI-derived measurements of human brain volumes from anatomical scans provide novel insights into normal and abnormal neuroanatomy. However, traditionally, these scans take 6-10 minutes to acquire, sometimes repeated in case of, or to preclude, subject motion. We have investigated shorter scans employing lower resolution and/or higher in-plane acceleration which have often yielded comparable morphometric results to conventional scans1,2,3. However, in-plane acceleration (GRAPPA) has decreasing time-benefits above 4-fold acceleration, while drastically reducing image SNR due to g-factor noise amplification. One approach to high acceleration in recent times has been incoherent undersampling with compressed-sensing reconstruction4. Here, we validate the morphometrics from rapid compressed-sensing MPRAGE scans5 to conventional MPRAGE scan acquisition parameters.Methods
All measurements were performed using a 3.0 T MRI scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany). Six subjects (mean: 29.3 years, 4 female) were scanned with a product 64-channel head coil. Each session included a conventional MPRAGE scan (MPR) acquired with recommended FreeSurfer parameters (6:02 min, TR/TI/TE=2530/1100/1.9 ms, matrix 256×256×176, resolution=1mm, GRAPPA R=2). Three scans were acquired using a prototype compressed-sensing MPRAGE sequence (csMPR) with 1mm isotropic resolution and varying acceleration levels (TR/TI/TE=2300/900/2.9 ms, matrix 256×240×176, acceleration factors of x4, x5 and x6, yielding scan times of 1:32, 1:14 and 1:02 min). One additional csMPR scan used parameters with 0.8mm spatial resolution (TE 3.0ms, matrix 320×300×224, acceleration = x4, scan time 2:25 min). All scans were acquired twice in the session. Images were first analyzed using the MRIQC package6 for QC metrics, and then using FreeSurfer v6.07 after all the scans from each subject were aligned using the FreeSurfer robust registration tool8. An automated parcellation of the cortex, subcortical and white matter structures was performed. Total gray and white-matter volume and average cortical thickness were determined.Results
CsMPR scans play out the same wave-forms as a conventional MPR – subjects experienced no discomfort or scanner-induced effects. Visually, the csMPR scans did exhibit lower SNR than the MPR, however, when evaluated computationally, this effect is limited to the white matter (see Fig. 1). Within the gray matter (GM) mask, the SNR was similar for the MPR and all 1mm resolution csMPR scans, regardless of undersampling/acceleration factor. Any variation was on the order of the test-retest uncertainty. The increase in SNR at 0.8mm resolution is due to the regularization parameter used in the CS reconstruction. White matter (WM) did show a ~ 20% decrease in SNR from MPR to the x4 csMPR scan, and further slight decreases with acceleration factor. However, our previous work has shown that morphometric values are robust to lower SNR than is conventionally obtained1-3. As a result, surface delineation and segmentation is highly reproducible between the MPR and csMPR scans. Fig. 2 shows an axial slice from an example subject after Freesurfer analysis with segmented sub-cortical structures and cortical regions highlighted in color. Agreement is very high. Fig. 3 shows total GM and WM volume, and Fig. 4 plots mean cortical thickness for 3 subjects across scan types. GM and WM volume showed internal consistency across scan type and acceleration. Increases in cortical thickness for all csMPR scans of ~ 0.05 – 0.1 mm are observed compared to the MPR scan, but otherwise variations are on the order of test-retest uncertainty.Discussion
Compressed-sensing MPRAGE offers the potential for significant acceleration with minimal impact on scanner hardware and/or image artifacts. As the impact of motion on anatomical scan quality and reliability of calculated morphometrics comes to be better understood9, an easy solution for making such 3D volumetric scans more immune to motion effects is to simply make them considerably shorter. We have demonstrated that csMPR scans be accelerated up to 6 times faster than a conventional MPR scan with minimal impact on bulk morphometrics. However, the incoherent undersampling approach and compressed-sensing reconstruction provide additional parameters beyond those available to conventional 3D cartesian imaging methods. Specifically, it is possible to tailor the number of k-space rows per TR during acquisition, which can impact the scan time and image quality; along with the regularizaton which controls image smoothness and SNR. As the SNR can be set arbitrarily in the reconstruction process, comparisons of SNR to conventional scans, and between levels of CS acceleration, should be made with caution. In optimizing csMPR for widespread use, a good compromise between noise and image sharpness must be found. Indeed, morphometry may prove to be the perfect way of assessing this. These parameters will be investigated further to optimize them for rapid T1-weighted images with sufficient SNR while not impacting morphometrics.Acknowledgements
Harvard Center for Brain Science; NIH Shared Instrumentation Grant S10OD020039; NIH Grants P50-MH106435, P41-RR14075, U24-RR021382.References
1. Holmes, A.J. et al., ‘Brain Genomics Superstruct Project initial data release with structural, functional, and behavioral measures’, Sci. Data 2015;2:150031.
2. Mair, R.W. et al., ‘Quantitative Comparison of Morphometric Data from Multi-Echo MPRAGE with Variable Acceleration and Different Head Coils’, Proceedings ISMRM 2013;21:947.
3. Mair, R.W. et al., ‘Impact of acquisition parameters on cortical thickness and volume derived from Multi-Echo MPRAGE scans’, Proceedings ISMRM 2016;24:1158.
4. Lustig M, et al., ‘Sparse MRI: The application of compressed sensing for rapid MR imaging’ Magn. Reson. Med.2007;58:1182-1195.
5. Mussard E. et al., ‘Accelerated MP2RAGE Imaging Using Sparse Iterative Reconstruction’, Proceedings ISMRM 2016;24:4216.
6. Esteban O. et al., ‘MRIQC: Advancing the Automatic Prediction of Image Quality in MRI from Unseen Sites’, PLOS ONE 2017;12:e0184661.
7. Fischl, B, ‘Measuring the thickness of the human cerebral cortex from magnetic resonance images’, Proc. Natl. Acad. Sci. USA, 2000;97:11050-11055.
8. Reuter, M., ‘Highly Accurate Inverse Consistent Registration: A Robust Approach’, NeuroImage, 2010;53;1181-1196.
9. Tisdall, M.D et al., ‘Prospective motion correction with volumetric navigators (vNavs) reduces the bias and variance in brain morphometry induced by subject motion’, Neuroimage,2016;127:11-22.
Figures
