Pushing the limits of in vivo diffusion MRI for the Human Connectome Project
Introduction
Following years of steady growth, diffusion MRI and fMRI have reached technological turning points in their respective mappings of human structural and functional connectivity, as have the computational tools to organize and share the resulting data. As part of the US National Institutes of Health's Blueprint Initiative for Neuroscience Research; Human Connectome Project (HCP), the MGH-UCLA collaboration was charged with pushing the frontier of extant acquisition technology with the goal of building dramatically more comprehensive and complete models of the structural Connectome than currently available. To achieve this we set out with a blank sheet of paper and asked; “what would an MR scanner optimized specifically for connectomics look like and what are the potential imaging improvements?” The result required developing and validating advances in every domain of MR technology except the magnet. First and foremost, we attempted to construct the highest performance gradient set ever attempted for human imaging. The resulting gradients utilize a peak gradient strength of 300 mT/m and are capable of slewing at 200 T/m/s. Although peripheral nerve stimulation (PNS) prevents the application of waveforms combining both high strength and high slew rate, the diffusion sequence requires primarily high strength alone during the diffusion encode and high slew rate at moderate Gmax (~ 40–50 mT/m) during the EPI readout. The goal was twofold; to shorten the TE of the spin echo diffusion, reducing signal losses from T2 processes, and to decrease the diffusion time Δ. Shortening TE has a simply characterized exponential effect on image sensitivity. The effect of the shortened diffusion time is less easily characterized but is expected to sharpen features in the water probability distribution function. Because the goal is efficient human diffusion imaging, the gradients need to perform at high duty-cycle for hours at a time without inefficient cool-down periods.
The sequence was re-examined with a goal of improving the notoriously in-efficient 2D spin echo diffusion sequence. Such 2D sequences become inefficient when TR becomes higher than the theoretical optimum; TR ≈ 1.2 T1 (about 1 s for white matter at 3 T), due to the number of slices needed to cover the head with 2D EPI. Thus typical diffusion acquisitions, with a TR of up to 10 s for a 2 mm isotropic acquisition, are enormously inefficient. The nuclear magnetization spends most of its time at equilibrium waiting to be sampled. The speed up was achieved by enabling a method of exciting and reading-out multiple slices simultaneously and separating them with parallel imaging methods. Although implemented a decade ago (Larkman et al., 2001), the method was not suitable for diffusion imaging since its g-factor penalty outweighs the envisioned efficiency benefit. Nonetheless, this technique was successfulness employed to significantly accelerate fMRI acquisition at 7T, where physiological fluctuation dominates the noise time-series (Moeller et al., 2008, Moeller et al., 2010). A significant reduction of the g-factor penalty in EPI was achieved by implementing a blipped CAIPIRHINA FOV shifting scheme, (Setsompop et al., 2012) thereby allowing a net gain in efficiency as well as an increase in the number of diffusion directions obtainable in a given scan time. Finally, the RF detection was augmented with a 64-channel brain array coil designed to increase parallel imaging performance (Keil et al., 2012). Taken together, the expected efficiency gain is nearly an order of magnitude for high b-value diffusion (3.5 × from shorter TE, 1.7 × from the Rslice = 3 SMS, and 1.4 fold from the RF coil).
In this paper, we describe the strategy and implementation of the MGH-UCLA Connectome scanner, our initial validation of its performance in SNR, sequence and diffusion ODF metrics, as well as a characterization of some of the concomitant problems of the methodology; such as peripheral and central nervous system stimulation, increased eddy current fields and gradient Maxwell terms.
Section snippets
Choice of B0 field strength
The first choice for the proposed scanner involved selecting the B0 field strength. Excellent quality EPI encoding, including high resolution single-shot, has been achieved at 7 T using highly parallel arrays coils by many groups, including ours. This immediately suggests that EPI-based methods such as diffusion imaging could potentially benefit from the 2 × increase in image (thermal) SNR observed in going from 3 T to 7 T (Triantafyllou et al., 2005, Vaughan et al., 2001). However, several issues
Gradient design and implementation
Many previous approaches to high strength gradients have relied on asymmetric head gradient designs. These designs are appealing since the small size creates an intrinsically efficient design (high Gmax per Ampere of current) as well as reduced PNS stimulation since the shoulders and torso are outside of the main gradient field. However, the asymmetric head gradient also has several negative features. The small size exacerbates the heating problem by reducing the mass and the volume available
Nerve stimulation thresholds
After successful fabrication, the gradient coil was tested for PNS at the factory using a standard threshold detection protocol in 35 healthy adult subjects (20 males 15 females). The gradient coil was energized in a laboratory setting (outside of the magnet) with the subjects supine with their head at isocenter. The PNS study was conducted according to the requirements defined by IEC 60601-2-33. chapter “51.105.1 Direct determination of the limits of the Gradient Output”. The gradients were
Eddy current effects and Concomitant terms (Maxwell's terms)
Both eddy current fields and concomitant fields are increased in stronger gradients. Although careful attention was paid to the shield coil design to successfully lower the eddy current fields as expressed as a percentage of the total gradient fields, the 7.5 fold higher gradient strength yielded eddy currents that were between 2 and 3 fold higher in absolute terms. Usually the first line of defense is to use a twice refocused spin echo sequence designed to eliminate the principle eddy current
Acoustic noise and vibrations
While a 7 fold increased gradient strength might potentially be accompanied by a similar increase in acoustic noise and vibrations, it appears that the effect of the thicker, heavier gradient, as well as increased torque and force balancing afforded by the larger volume, offset this problem. The gradient running at 300 mT/m produces a lower acoustic noise than the conventional 40 mT/m 3 T and qualitative assessment of the vibration suggests it is also reduced.
Achieved TE and diffusion time (Δ) reductions and SNR gains
Fig. 5 shows the minimum TE obtained for a standard Stejskal–Tanner spin echo diffusion sequence as a function of b value for a 2 mm isotropic EPI readout (200 FOV, 3/4 partial Fourier, and R = 2 acceleration) as a function of maximum gradient strength. The TEmin is shown for a gradient strength of 40 mT/m, 80 mT/m, 100 mT/m and 300 mT/m. The largest reduction in TE and Δ occurs for the higher b values. For Gmax = 40, 80, 100 and 300 mT/m and b = 10,000 s/mm2, the diffusion time (Δ) was; 56.4, 38.4, 33.9,
RF design and implementation
Since parallel image reconstruction (both conventional in-plane and in the slice directions) was central to our acquisition strategy, we set out to create a state-of-the-art highly parallel RF brain coil. (Keil et al., 2012) The resulting 64 channel brain array is shown in Fig. 7. The array former splits into two halves and yet retains the critical overlap of the circular elements (~ 55–65 mm dia). The array adheres to the head shape to a degree previously not attempted. It is also mounted
Simultaneous multi-Slice with blipped-CAIPI FOV shift
The diffusion encoding pulses encode the water displacement in the whole brain, but are followed by the readout of only a single slice. This wastes a significant fraction of the imaging time. An alternative view is that a given spin is only sampled for a small fraction of the TR period; an inefficiency which grows with the number of 2D slices. The ultra high strength gradients allow for a significant shortening of the diffusion encoding time, while the use of parallel imaging to reduce the
Susceptibility distortion mitigation
The geometric image distortion induced by susceptibility gradients in the phase encode direction of EPI remains a significant problem for single shot SE-EPI based diffusion acquisitions. While the hardware advances described do not fully mitigate this problem, they reduce these distortions in two ways. Firstly, the stronger gradients allow a faster EPI readout than conventional gradients even though PNS prevents the use of the full gradient strength during the EPI readout. Nonetheless the
Diffusion imaging evaluation
While the higher q-space encoding, in principle, will encode higher detail in the PDF and therefore the Orientation Distribution Function (ODF), it does so at reduced SNR due to the high b-value. We assess the benefit of the higher gradient strengths in Fig. 13 by comparing the first and second fiber uncertainty metrics in b = 10,000 s/mm2 q-ball acquisitions acquired on the Connectome scanner using Gmax = 40 mT/m, 100 mT/m and 300 mT/m. The acquisition used the single refocused spin echo sequences, TR =
Conclusion
Gradient strength, acquisition efficiency and detection sensitivity are critical determinants of sensitivity in diffusion imaging, especially for high b-value high angular resolution acquisitions. Encouraged by the knowledge that small bore scanners with high gradient strength appeared to produce a richer structural Connectome in fixed tissue than is possible in living humans, we set out to re-engineer a clinical scanner to match the small bore capabilities. Owing to synergistic effects of
Acknowledgments
This work acknowledges funding from the National Institutes of Health, NIH Blueprint Initiative for Neuroscience Research grant U01MH093765, National Institutes of Health grant P41EB015896, and NIH NIBIB grant K99/R00EB012107.
Note that the MGH-UCLA Connectome scanner is a Work in Progress. The information about this product is preliminary. The product is under development and is not commercially available in the U.S. and its future availability cannot be assured.
Conflict of interest
The authors
References (77)
- et al.
Deconvolution in diffusion, spectrum imaging
NeuroImage
(2010) - et al.
Ultra-fast MRI of the human brain with simultaneous multi-slice imaging
J. Magn. Reson.
(2013) - et al.
Spherical wavelet transform for ODF sharpening
Med. Image Anal.
(2010) - et al.
Probabilistic identification and estimation of noise (PIESNO): a self-consistent approach and its applications in MRI
J. Magn. Reson.
(2009) - et al.
Development and initial evaluation of 7-T q-ball imaging of the human brain
Magn. Reson. Imaging
(2008) - et al.
Resolution of complex tissue microarchitecture using the diffusion orientation transform (DOT)
NeuroImage
(2006) - et al.
Retinal origin of phosphenes to transcranial alternating current stimulation
Clin. Neurophysiol.
(2010) - et al.
Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution
NeuroImage
(2004) - et al.
Resolving crossing fibres using constrained spherical deconvolution: validation using diffusion-weighted imaging phantom data
NeuroImage
(2008) - et al.
Comparison of physiological noise at 1.5 T, 3 T and 7 T and optimization of fMRI acquisition parameters
NeuroImage
(2005)
Multiple Q-shell ODF reconstruction in Q-ball imaging
Med. Image Comput. Comput. Assist. Interv.
Degenerate mode band-pass birdcage coil for accelerated parallel excitation
Magn. Reson. Med.
Delta relaxation enhanced MR: improving activation-specificity of molecular probes through R1 dispersion imaging
Magn. Reson. Med.
Measurement of fiber orientation distributions using high angular resolution diffusion imaging
Magn. Reson. Med.
A comprehensive gaussian process framework for correcting distortions and movements in diffusion images
Guidelines for limiting exposure to time-varying electric and magnetic fields (1 Hz to 100 kHz)
Health Phys.
Ch. 10 correction gradients
Handbook of MRI pulse sequences
Accelerated diffusion spectrum imaging with compressed sensing using adaptive dictionaries
Magn. Reson. Med.
Controlled aliasing in parallel imaging results in higher acceleration (CAIPIRINHA) for multi-slice imaging
Magn. Reson. Med.
Array compression for MRI with large coil arrays
Magn. Reson. Med.
Mathematical description of q-space in spherical coordinates: exact q-ball imaging
Magn. Reson. Med.
Quality assessment of high angular resolution diffusion imaging data using bootstrap on Q-ball reconstruction
J. Magn. Reson. Imaging
International standard IEC 60601-2-33 medical electrical equipment
Particular Requirements for the Safety of Magnetic Resonance Equipment for Medical Diagnosis
Signal-to-noise measurements in magnitude images from NMR phased arrays
Magn. Reson. Med.
Measuring T2 and T2′ in the brain at 1.5 T, 3 T and 7 T using a hybrid gradient echo-spin echo sequence and EPI
Signal-to-noise ratio and parallel imaging performance of a 16-channel receive-only brain coil array at 3.0 Tesla
Magn. Reson. Med.
A model-based deconvolution approach to solve fiber crossing in diffusion-weighted MR imaging
IEEE Trans. Biomed. Eng.
Regularized, fast, and robust analytical Q-ball imaging
Magn. Reson. Med.
Deterministic and probabilistic tractography based on complex fibre orientation distributions
IEEE Trans. Med. Imaging
Simultaneous echo refocusing in EPI
Magn. Reson. Med.
Multiplexed echo planar imaging for sub-second whole brain FMRI and fast diffusion imaging
PLoS One
Sparse DSI: Learning DSIstructure for denoising and fast imaging
Med. Image Comput. Comput. Assist. Interv.
SAFE Model — a new method for predicting peripheral nerve stimulations in MRI
Isotropic submillimeter fMRI in the human brain at 7 T: combining reduced field-of-view imaging and partially parallel acquisitions
Magn. Reson. Med.
Parallel imaging in non-bijective, curvilinear magnetic field gradients: a concept study
MAGMA
Q-ball reconstruction of multimodal fiber orientations using the spherical harmonic basis
Magn. Reson. Med.
Squashing peanuts and smashing pumpkins: how noise distorts diffusion-weighted MR data
Magn. Reson. Med.
A 64-channel 3 T array coil for accelerated brain MRI
Magn. Reson. Med.
Cited by (424)
Progressive dual-domain-transfer cycleGAN for unsupervised MRI reconstruction
2024, NeurocomputingEfficient approximate signal reconstruction for correction of gradient nonlinearities in diffusion-weighted imaging
2023, Magnetic Resonance Imaging