A Time-Efficient Acquisition Protocol For Multi-Purpose Diffusion‑Weighted Microstructural Imaging At 7T
Farshid Sepehrband1,2, Kieran O’Brien1,3, and Markus Barth1

1Centre for Advanced Imaging, University of Queensland, Brisbane, Australia, 2Laboratory of Neuro Imaging, USC Mark and Mary Stevens Neuroimaging and Informatics Institute, Keck School of Medicine of USC, Los Angeles, CA, United States, 3Siemens Healthcare Pty Ltd, Brisbane, Australia


Several diffusion-weighted MRI techniques for modeling tissue microstructure have been developed and validated during the past two decades. While offering various neuroanatomical inferences, these techniques differ in their proposed optimal acquisition design, which impede clinicians and researchers to benefit from all potential inference methods, particularly when limited time is available. We examined the performance of the most common diffusion models with respect to acquisition parameters at 7T when limiting the acquisition time to about 10 minutes. The most balanced compromise among all combinations in terms of the robustness of the estimates was a two-shell scheme with b-values of 1,000 and 2,500 s/mm2 with 75 diffusion-encoding gradients, 25 and 50 samples for low and high b-values, respectively.


To obtain a balanced DWI acquisition scheme that can cover the varying requirements of the most commonly used microstructural models at 7T.


We acquired a reference dataset with multiple shells, covering a wide range of b‑values (700 to 3,000 s/mm2) using an optimal multi-shell sampling scheme1 that includes the suggested protocols of the microstructural DWI techniques listed below. From this data set we generated a large number of sub‑datasets from the multi-shell data (around 900 combinations). The sub-datasets were evaluated based on different criteria for the following techniques: NODDI (neurite orientation distribution and density index), CSD (constrained spherical deconvolution), DTI (diffusion tensor imaging) and WMTI (white matter tract integrity) 2-5. For NODDI the entire multi-shell data was concatenated and the NODDI tissue model was fitted to the complete reference data set. The derived model parameters were used as the gold standard to evaluate parameters obtained from the various sub-datasets. For CSD the linear fascicle evaluation6 technique and fibre reconstruction error rate compared with the full dataset were used to assess b-values ≥2000s/mm2. For DTI the root mean square errors (RMSE) of sub-datasets for b-values ≤1300 s/mm2 were compared to their full datasets. Finally, the optimal choices from above were compared with the suggested WMTI protocol2: 1,000 and 2,000 s/mm2. Using these analysis results from all methods we defined an optimal range by identifying protocols that are not particularly suitable for multi-purpose diffusion-weighted imaging (Figure 5).


A healthy, 20-year-old right-handed male was scanned after giving written informed consent according to the local ethics committee approval. The acquisition was performed on a 7T whole-body research scanner (Siemens Healthcare, Erlangen, Germany) with a maximum gradient of 70 mT/m and a maximum slew rate of 200 T/m/s using a single-channel quadrature transmit RF coil and a 32-channel receive array coil (Nova Medical Inc, MA). The protocol consisted of six high angular resolution diffusion-weighted shells, with b-values of 700, 1000, 1300, 2000, 2500 and 3000 s/mm2, respectively, using Stejskal-Tanner DW gradients. Other parameters: 1.8 mm isotropic voxels, 70 slices, 120×120 matrix size, field-of-view 216×216mm, phase partial Fourier 6/8, GRAPPA acceleration factor 3. A fixed TE of 68.4ms and a TR of 8100 ms were used with an effective echo spacing of 0.22ms. Total scan time was 38m:58s for the complete reference data set. A sampling scheme consisting of 90 directions with a low:high b-value ratio of 1:2 were designed using the q-space sampling web application1.


NODDI: The six combinations that had the highest correlation with the gold standard are plotted. For intra-cellular volume fraction (ICVF), the 700-2500 and 1000-2500 dataset (first-second b-values) had the highest correlation with the gold standard (Figure 1). The orientation distribution index (ODI) results showed a high similarity (correlation >0.9) across all datasets. Combinations with either a b-values of 1300 and 3000 or combinations with ≤60 samples underperformed and consequently were rejected by NODDI analysis (Figure 5). CSD: No decrease in the mean prediction errors as a result of b-value were observed, rather a slight increase was apparent (Figure 2A). This slight increase could be due to the lower SNR of datasets with high b-values. The number of samples required to meet a fibre reconstruction error equivalent to the full data set was ≥46, ≥48, and ≥50 for b-values of 2000, 2500 and 3000 s/mm2, respectively (Figure 2B-D). DTI: The RMSE showed little improvement after 20 samples for b-values of 1000 and 1300 s/mm2. For a b‑value of 700 s/mm2 more samples were required (≥24) to obtain the acceptable relative error rate compared to the full dataset (Figure 3). WMTI: the evaluation of the 1000-2500 and 1000-3000 datasets (Figure 4) showed that the correlation with the estimates from the conventional protocol decreased when the second b-value was larger.


Overall, our suggested design for a multi-purpose diffusion-weighted microstructural imaging at 7T with a minimal number of total samples that best satisfies all models is a protocol with b-values of 1000 and 2500 s/mm2, with 25 and 50 samples, uniformly distributed over two shells. With the 1.8 mm isotropic resolution, TR of 8100 ms and 5 unweighted measurements (b0), this acquisition scheme could be scanned in <11mins at 7T. The acquired data can be used for DTI, tractography, NODDI, WMTI and potentially more techniques.


MB acknowledges funding from ARC Future Fellowship grant FT140100865. The authors acknowledge the facilities of the National Imaging Facility at the Centre for Advanced Imaging, University of Queensland and the scientific support of Siemens Ltd, Bowen Hills, Australia.


1. Caruyer E, Lenglet C, Sapiro G, Deriche R. Design of multishell sampling schemes with uniform coverage in diffusion MRI. Magn. Reson. Med. 2013;69:1534–1540. doi: 10.1002/mrm.24736.

2. Fieremans E, Jensen JH, Helpern JA. White matter characterization with diffusional kurtosis imaging. Neuroimage 2011;58:177–188. doi: doi: 10.1016/j.neuroimage.2011.06.006.

3. Zhang H, Schneider T, Wheeler-Kingshott CA, Alexander DC. NODDI: practical in vivo neurite orientation dispersion and density imaging of the human brain. Neuroimage 2012;61:1000–1016. doi: 10.1016/j.neuroimage.2012.03.072.

4. Basser PJ, Mattiello J, LeBihan D. MR diffusion tensor spectroscopy and imaging. Biophysical Journal 1994;66:259–267. doi: 10.1016/S0006-3495(94)80775-1.

5. Tournier J-D, Yeh C-H, Calamante F, Cho K-H, Connelly A, Lin C-P. Resolving crossing fibres using constrained spherical deconvolution: Validation using diffusion-weighted imaging phantom data. Neuroimage 2007;42:617–625. doi: 10.1016/j.neuroimage.2008.05.002.

6. Pestilli F, Yeatman JD, Rokem A, Kay KN, Wandell BA. Evaluation and statistical inference for human connectomes. Nat Methods 2014;11:1058–1063. doi: 10.1038/nmeth.3098.


Figure 1. Assessment of NODDI-derived metrics, obtained from different sub-datasets, when compared against the gold standard, i.e. concatenation of all acquired DW images. Correlation of ICVF (A) and ODI (B) metrics with the gold standard for b-value combinations with highest ranking, across different sampling numbers.

Figure 2. Evaluation of fibre reconstruction results using (A) the LiFE technique6; CSD with a maximum spherical harmonic of order 6 was fitted to all voxels of white matter. Mean, standard deviation and distribution of RMSE values are shown. (B-D) shows RMSEs of fibre-count measures against full datasets.

Figure 3. Evaluating DTI metrics obtained from different b-values and samples. Plots show RMSE of DTI-derived maps, between different sub-datasets and their corresponding full dataset. The sudden decrease in RMSE occurs when an additional b0 data set is added. Errors unit is mm2/ms.

Figure 4. Comparing 1000-2500 and 1000-3000 datasets with the one from the conventional protocol of WMTI (1000-2000). A total of 72 samples was used. Note that 1000-2500 dataset has a much higher correlation with the conventional protocol compared to the 1000-3000 dataset. Diffusivities units are μm2/ms.

Figure 5. Summary of the analysis. All combinations of b-values and sampling numbers are shown and shaded based on our analysis (e.g. b-value of 2500 s/mm2 with 36 DW samples was rejected once by the outcomes of NODDI analysis and once by the outcomes of CSD analysis).

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