Distortion-matched T1-maps and bias-corrected T1w-images as anatomical reference for submillimeter-resolution fMRI
Wietske van der Zwaag1, Pieter Buur1, Maarten Versluis2, and José P. Marques3

1Spinoza Centre for Neuroimaging, Amsterdam, Netherlands, 2Philips Healthcare, Best, Netherlands, 3Donders Institute for Brain, Cognition and Behaviour, Nijmegen, Netherlands

Synopsis

Achieving sufficiently good quality co-registration between the anatomical and functional images is currently a large stumbling block for laminar fMRI. Here, we present a distortion-matched T1weighted/T1-estimation mapping approach using two 3D-EPI readouts per inversion, following the MP2RAGE signal combination. 0.7mm isotropic T1 data with matching distortions to a 0.7mm isotropic fMRI protocol can be acquired in less than two minutes.

Target audience

Anyone working on laminar fMRI

Purpose

To generate resolution and distortion-matched T1-maps and bias-field corrected T1w-images for use as anatomical reference to submillimetre fMRI.

The co-registration of high resolution functional data is problematic because of (small) local and global distortions which complicate alignment with the structural scans. The consequences of misalignment are further enhanced when trying to disentangle functional processes as a function of cortical depth. Cortical depth surfaces, laminae, are usually defined in the anatomical dataset and subsequently overlaid on the functional data from which the depth-dependent signals are derived. Hence, co-registration needs to be correct, locally, with a precision of around 0.1mm.

Here, the two GRE readouts of the MP2RAGE were replaced with two 3D-EPI blocks, with equal length readouts as the functional data, to obtain T1 estimation maps and bias-field corrected T1w images with distortions matched to those of the functional data.

Methods

A 0.7mm isotropic 3D-EPI protocol(1) for submillimetre fMRI was adapted to generate the T1-estimation maps and bias-field corrected T1w-images using the Multiple Interleaved Scanning Sequences, MISS, environment on the Philips platform. The 3D-EPI parameters were: FOV 120 x131 x 24 mm, matrix size 132 x 182 x 34, TR/TE = 57/28ms, EPI factor = 27, SENSE undersampling factor 3.5(LR)*1.3(AP), 72 EPI readouts (segments) per volume, volume acquisition time 4.0s.

The MP2RAGE sequence(2) contains two gradient echo readout blocks following a single inversion. Signal from both images is combined following: $$$ MP2RAGE = \frac{GRE_{TI1} GRE_{TI2}}{GRE_{TI1}^{2} +GRE_{TI2}^{2}} $$$. In the 3D-EPI equivalent, T1-imaging with 2 3D-EPI’s, or T123DEPI, following a single inversion pulse a much larger k-space section is acquired (up to half or a quarter for this 3D-EPI protocol), drastically reducing acquisition time. A sequence diagram is shown in Figure 1. Simulations following(2) were used to derive protocol parameters for the MP2RAGE and T123DEPI protocols.

Four volunteers were scanned at 7T (Philips, Netherlands) with a 32-channel surface coil (MRcoils, Netherlands, n=2) or a 32-channel volume coil (Nova Medical, USA, n=2). Two T123DEPI protocols were compared, both matching the FOV and matrix sizes of the fMRI acquisitions:

1) 36 segments per readout (2 blocks), TRT123DEPI =8.25s, TI1/TI2=1200/3800ms, α1/ α2 = 14/10, 4 averages, total acquisition time 80s.

2) 18 segments per readout (4 blocks), TRT123DEPI =10s, TI1/TI2=1000/2700ms, α1/ α2 = 20/16, 4 averages, total acquisition time 128s.

MP2RAGE anatomical data were acquired for 3 volunteers (2x volume coil, 1x surface coil): voxel size 0.64 mm isotropic, FOV 205 x 205 x 164 mm, TRMP2RAGE/TE/TI1/TI2 = 8s/6.2/800/2700 ms, α1/ α2 = 7/5, total acquisition time 11min.

For one volume-coil volunteer, 30 volumes of resting 3D-EPI were also acquired. For the fourth volunteer, a 6-min functional run with the corresponding 3D-EPI acquisition was acquired with a visual stimulus and alternating ON=4s, OFF=10s blocks. Functional data were analysed with SPM12 (GLM, 1mm smoothing).

Results and Discussion

The division of the 3DEPI readout in 4 clustered blocks means an optimal TI1 can be used (Figure 1), whereas in the 2-block protocol the minimum possible TI1 was 1200ms. Both protocols yielded homogeneous T1-weighting, comparable to the MP2RAGE, throughout the acquisition slab (Figure 2). Although data was combined over a large period of time (TRT123DEPI*Nblocks = 8.25*4 s instead of 10*2), 4-block data did not demonstrate more motion sensitivity than the 2-block data. Generally, SNR was higher in the 4block T123DEPI (see Figure 2), in agreement with the simulations results.

The 11-minute MP2RAGE acquisition yielded much superior SNR to the T123DEPI’s, but did not coregister with the same local fidelity as the T123DEPI to the mean EPI (Figure 3). Overall coregistration was successful for all 3 T1w-images (left panels Figure 3), but when the gray matter boundaries of the MP2RAGE are shown overlaid on the 3D-EPI data, small mis-registrations appear due to local distortions; around CSF/veins in de sulci (light blue arrow); on the cortical surface (mid blue arrow) and on the gray-white matter boundary (dark blue arrow). Distortions in the T123DEPI match those of the mean 3D-EPI in the top right panel.

Gray matter T1-values were highly reproducible between MP2RAGE and the two T123DEPI protocols (Figure 4). Functional activation perfectly aligned with cortical gray matter and/or pial veins (Figure 5). Generally, SNR could easily be improved by acquiring more averages given the very short acquisition times used here.

Conclusion

With the increased interest in laminar fMRI, high confidence in lamina definition is becoming more and more important. The iso-resolution, distortion-matched T1w-images and T1–maps can provide anatomical reference data which is straightforward to co-register to the functional 3D-EPI data and can be acquired in less than two minutes.

Acknowledgements

The authors would like to thank Dr Kamil Ugurbil for emphasizing the need for a distortion-matched T1w-EPI at the laminar fMRI meeting in Nijmegen.

References

1) Petridou et al, 2013, NMR Biomedicine

2) Marques et al, 2010, Neuroimage

Figures

Figure 1. Sequence diagram of the T123DEPI. Two 3D-EPI blocks follow each inversion. For Nblocks = 2, half of all EPI readouts are acquired in one 3D-EPI block. For Nblocks = 4, a quarter.

Figure 2. Example slices from the 0.7mm3 T1w-T123DEPI images and comparable slices from the 0.7mm3 MP2RAGE. All data were skull-stripped and the same intensity range was used for all. Note the homogeneous contrast throughout the slices, in both the volume and surface coil data. 2-block data have slightly lower T1 contrast due to the longer TI1, but 4-block T1w-T123DEPI images have near-identical contrast to the MP2RAGE standard.

Figure 3. SPM coregistration results for T1w-images from an MP2RAGE and two T123DEPI acquisitions to a mean EPI volume (n=30). Right panels display the area marked with a white box (top left panel). Red and green contours indicate the white/gray matter boundary and gray matter/CSF boundary in the MP2RAGE. Blue arrows indicate regions where the MP2RAGE boundaries do not match the anatomical structure seen in the mean EPI.

Figure 4. Example slices from the 0.7mm3 T1-maps from both T123DEPI acquisitions and the MP2RAGE, corresponding to the same slices as shown in Figure 2. The same brain masks were used for the T1-maps as for the T1w-images. All maps are scaled from 0-3000 ms. Gray matter T1-values were highly reproducible between all 3 protocols.

Figure 5. Five consecutive example slices from the activation map overlaid on the corresponding brain-masked T123DEPI T1w-image. (p<0.05). The T123DEPI T1w-image was co-registered to the mean EPI using a rigid-body co-registration in SPM. Images were cropped for display purposes. Note the excellent correspondence between location of BOLD responses and the T1w-image structures.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0071