Introduction

MRI at 7T is a promising neuroimaging tool with great potential clinical yield, e.g., in detection and delineation of brain lesions [1]. By the virtue of submillimeter spatial resolution, increased susceptibility contrast and SNR, standard structural sequences and contrast weightings may provide increased sensitivity during both visual inspection and computational methods. But 7T MRI is also challenged by increased B₁⁺-induced spatial inhomogeneity [1,2], and 2D-based RF gain optimization may not be adequate for dealing with the large B₁ field inhomogeneity variations encountered at ultra-high field. Sub-optimal RF gain optimization may cause additional B₁⁺ inhomogeneity, which may hamper visual inspection, and even after bias-field correction leave images unusable for computational analyses. Some sequences, e.g., MP2RAGE [2], were developed specifically to mitigate these B₁-induced image variations. However, such sequences may have other shortcomings, e.g., prolonged scan time, not being automatically reconstructed at the scanner, or a different contrast than what a dedicated neuroradiologist at the local center prefers. It is therefore desirable to seek out stable solutions to this commonly encountered issue at 7T MRI with a classical setup. A simple and widely used method is to apply dielectric pads on each side of the head. But even with these, the spatial dependency of the B₁⁺-field may induce signal dropout in mesial or lateral brain regions (figure 1), which may prevent optimal conditions for inspection or analyses. Such signal dropout and inhomogeneity may further vary with head size, positioning, image contrast and sequence type, and leaves clinicians and researchers with a highly varying image quality. We investigated the effects of combined use of large bilateral dielectric pads and 3D-based RF gain optimization on the stability of bias fields in 3D MPRAGE images. To compare across cohorts, we used the coefficient of variation of the computed bias field generated in post-processing of 3D MPRAGE scans.

Methods

Informed consent was obtained according to local ethical guidelines, in relation to a larger scan protocol. 3D MPRAGE images (0.7 mm isotropic) were acquired on an actively shielded 7T MR system (Philips, Achieva, Best, The Netherlands), with a quadrature 32/2 Rx/Tx coil (Nova Medical, Wilmington, MA), and dielectric pads on both sides of the head.
Ninety-five subjects were included, and divided into three groups: Group 1 (n=48) with smaller dielectric pads and 2D-based RF gain optimization, group 2 (n=29) with renewed, larger pads (19x19 cm, Multiwave Imaging, Marseille, France) and 2D-based RF gain optimization, and group 3 (n=18) with the renewed pads and 3D-based RF gain optimization. This was done by including a DREAM sequence to map the B₁⁺ field (3.5mm x 3.5mm in-plane resolution, 36 slices, FOV: 240 mm x 180 mm x 240 mm, STEAM preparation flip angle: 35 deg.). In general, we observed ~40% variation in the B₁⁺-maps, and a tendency of the RF gain to be overestimated (see figure 1). To counter-act this, we implemented an automated adjustment of the RF gain, based on the acquired DREAM map: After light smoothing and masking, the maximum B₁⁺-intensity was identified and used for setting the RF gain. Bias fields were computed with bias-field correction of all MPRAGE images with step-size 3, FWHM 60 mm and light regularization (0.001) in SPM12. Subsequently, we computed a coefficient of variation (CV), defined as the ratio of the standard deviation to the mean of the bias-field intensity per axial slice inside a brainmask. We used ANOVA test to compare the per-subject average of CV (|CV|) between the three groups.

Results

As demonstrated in figure 2, there was a larger spread of CV across subjects in group 1, compared to group 2 and 3. There was a slight trend in increased spread of CV values in group 2 compared to group 3. Boxplots on figure 3 shows the distributions of |CV| in the three groups. ANOVA test revealed significant differences (p=1.7307e-21). This was largely driven by the variation in group 1, albeit group 2 and 3 was also found to be significantly different (p= 0.0387).

Discussion

By comparing CV across groups in our study, we saw significant differences between different approaches for minimizing B₁-induced bias fields. The combination of large pads and 3D-based RF gain optimization (group 3) yielded the smallest |CV|-variation, indicating increased stability in the bias field and thereby B₁⁺. However, this is a quantitative and indirect measure of the quality-parameter “stability”, and thus may not be translatable to all subjects, nor in all contrast weightings. While we analyzed 3D MPRAGE, other sequences, e.g., 3D FLAIR, are even more vulnerable to B₁⁺-induced variations, and would likely show larger effects. In our experience, the apparent quality improvement observed here are likewise seen for other sequences during radiological evaluation, and the implementation of the 3D-based RF gain optimization have completely mitigated cases of severe signal dropout in mesial brain regions across sequences.

Conclusions

When performing structural neuroimaging at 7T with a classical coil setup and transmit mode (i.e., low RF transmit channel count), we recommend using a combination of large dielectric pads on each side of the head along with 3D-based RF gain optimization

Acknowledgements

The project is supported by the Independent Research Fund Denmark. The 7T scanner was donated by the John and Birthe Meyer Foundation and The Danish Agency for Science, Technology and Innovation (grant no. 0601-01370B).