Philipp Ehses

Distortion-free high-resolution fMRI at 9.4 T

Introduction

Signal-to-noise-ratio (SNR) in MRI is expected to grow linearly with increasing field, while a BOLD increase of more than linear is expected [1], allowing a significant higher resolution in BOLD fMRI at ultra-high fields [2]. On the flip side, increased B₀ field inhomogeneities aggravate distortion artifacts in EPI BOLD imaging at higher field strengths. Additionally, stronger T₂^* relaxation during the EPI read-out leads to increased blurring effects, partially offsetting the higher nominal resolution. Both of these problems can be alleviated by shortening the EPI read-out using parallel imaging and/or segmentation. In the extreme case, only a single echo is acquired in every TR, transforming the EPI into a conventional gradient-echo (GRE) sequence. The aim of this work is to develop an effective single-echo GRE sequence for BOLD fMRI at ultra-high magnetic field. To this end, echo-shifting (ES) [3] was combined with an interleaved slice order, in order to optimize echo time, scan time, and SNR for multi-slice BOLD fMRI.

Methods

All experiments were performed at 9.4 T on a healthy volunteer with informed consent and IRB approval. A custom-built head coil was used for signal transmission/reception (8 transmit / 18 receive channels). An echo-shifted GRE with interleaved slices was developed, using an acquisition strategy that was recently proposed for an EPI sequence [4]. Parallel imaging (R=2 using GRAPPA [5] with ACS pre-scan) and partial Fourier (6/8) were used to accelerate image acquisition. Other sequence parameters were as follows: FA = 22°, effective TR/TE = 80/20 ms with ten interleaved slices and two echo-shifts, slice thickness = 1 mm (+25% slice distance), 1 mm in-plane resolution, TA per repetition = 4.32s, 70 repetitions, total TA = 5min. The paradigm consisted of finger tapping with the right hand, alternating between 20 s of rest and 20 s of tapping. For analysis, the data were processed with FSL FEAT [6,7] after brain extraction, using a standard hemodynamic response function and temporal filtering. Data processing was repeated with and without spatial smoothing, using a Gaussian filter kernel (FWHM=1mm). In addition, a high-resolution GRE image served as a reference (FA = 25°, TR/TE = 500/14ms, resolution = 0.23x0.23 mm², slice thickness = 1mm, 16 slices). Activation maps were registered to this reference using a six-parameter rigid body model.

Initial Results

Figure 1 shows the time series without spatial smoothing for the voxel with maximum z-score. A strong BOLD activation (±40%) is observed, explained by the fact that the corresponding voxel lies inside a vein. Typical activations in the motor cortex were approximately ±4%. Figure 2ab shows an overlay of the smoothed activation map over a) the averaged raw image and b) registered to the reference image. In Figure 2c-e, the left motor cortex is superimposed with activation maps with and without spatial smoothing. The position of veins close to the motor cortex strongly correlates with high BOLD activation.

Discussion/Conclusion

The results show that the proposed method is well suited for high-resolution fMRI at ultra-high magnetic field. Advantages compared to EPI are that it results in less T₂^* blurred and virtually distortion-free images. This comes at the cost of longer scan times (4.32 s for 10 slices vs. 2s for a typical >=20 slices EPI). Further optimizations and higher acceleration factors may help to close this gap, although we believe that EPI will remain more time-efficient. Another possible advantage for auditory fMRI applications is that the proposed sequence creates very monotonous, mostly low frequency noise. Interestingly, and apart from its advantages compared to EPI, the proposed method also becomes more feasible with higher field: Due to shorter T₂^*, less echo-shifts (i.e. less gradient switching and diffusion related signal loss) are required to optimize the echo time for BOLD.

References

1. Gati JS, Menon RS, Ugurbil K, Rutt BK. Experimental determination of the BOLD field strength dependence in vessels and tissue. Magn Reson Med 1997;38:296-302.
2. Speck O, Stadler J, Zaitsev M. High resolution single-shot EPI at 7T. MAGMA 2008;21:73-86.
3. Moonen CTW, Liu G, Van Gelderen P, Sobering G. A fast gradient-recalled MRI technique with increased sensitivity to dynamic susceptibility effects. Magn Reson Med. 1992;26:184–189.
4. Gibson A, Peters AM, Bowtell R. Echo-shifted multislice EPI for high-speed fMRI. Magn Reson Imaging. 2006;24:433-42.
5. Griswold MA, Jakob PM, Heidemann RM, Nittka M, Jellus V, Wang J, Kiefer B, Haase A. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magn Reson Med 2002;47:1202-10.
6. Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TEJ, Johansen-Berg H, Bannister PR, De Luca M, Drobnjak I, Flitney DE, Niazy RK, Saunders J, Vickers J, Zhang Y, De Stefano N, Brady JM, Matthewsa PM. Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage 2004;23:208–19.
7. Woolrich MW, Jbabdi S, Patenaude B, Chappell M, Makni S,  Behrens T, Beckmann C, Jenkinson M, Smith SM. Bayesian analysis of neuroimaging data in FSL. NeuroImage 2009;45:173-86.

Figure 1: Time series plot of the voxel with maximum z-score (=11.4) (without spatial smoothing). The very strong BOLD response can be explained by the fact that the voxel lies directly inside a vein.

Figure 2: a) Overlay of BOLD activation (smoothed with FWHM=1mm) over averaged raw image. b) Activation from a) registered to reference. c-e) Zoom in on left motor cortex (different slice than in ab), superimposed with d) registered activation map (FWHM=0) and e) registered and smoothed map (FWHM=1mm). Veins close to the motor cortex are indicated by white arrows.