Synopsis

Keywords: Data Acquisition, Brain

There are significant benefits to segmented 3D EPI fMRI acquisitions, which acquires high spatial and temporal resolution across the whole brain to better understand brain neuronal activity. This can be achieved through accelerated 3D EPI imaging, which provides rapid whole-brain coverage at the cost of SNR efficiency resulting from frame-by-frame reconstruction. Here, we utilize temporal information for whole-brain coverage on 8-fold accelerated 7T fMRI acquisition without altering the subsequent fMRI results.

Introduction

Ultra-high field increases fMRI signal-to-noise ratio (SNR) and sensitivity to BOLD contrast, making it possible to detect brain activation areas with sub-millimeter resolution, which has become the primary neuroscience research tool for mapping the brain activities across thin cortical layers [1-3]. Particularly, neuroscientists have developed an array of methods capable of looking at whole-brain functional activity and connectivity analysis [4-6].
The purpose of this study is to investigate the feasibility of whole-brain 3D EPI fMRI at 0.75mm and 0.64mm isotropic resolutions on the Next Generation 7T scanner using temporal random walk sampling pattern. Experimental studies confirm: 1) leveraging temporal information in fMRI reconstruction provides a high SNR efficiency across time and 2) the resulting high tSNR leads to increased BOLD activations.

Methods

Pulse Sequence and Sampling Pattern: We used multi-shot based segmented 3D EPI sequence for data acquisition to achieve reduced off-resonance artifacts and T2*-blurring, and shorter TEs. The 3D spatial encoding employed variable density CAIPIRINHA (VD-CAIPI) sampling pattern by segmenting k_z-space with variable width, in which the central k_z-space is sampled at the Nyquist rate while the outer k_z-space is regularly under-sampled based on the CAIPI pattern [7,8]. Additionally, as depicted in Figure 1, the time-wise spatial random blips allow extra spatial encoding across time in an incoherently complementary manner within the EPI framework while keeping the coherence on the spatial axes without changing the EPI blip size, resulting in random undersampled spatiotemporal data structure (Fig 1A).

Data Acquisition: All fMRI data were collected on the next generation 7T scanner (Siemens Healthcare, Erlangen, Germany) equipped with a 200 mT/m, 900 T/m/s gradient system and a 64-channel head coil. The imaging parameters are: FOV=144x189x132mm³, TR/TE=5s/23ms, FA=18^o, resolution=0.64mm and 0.75mm isotropic, number of segments=2, PF=6/8 (phase), and 8-fold acceleration (R_yxR_z=2x4). As a competing method, we used skipped CAIPI 3D EPI [9], and the image parameters were matched with our sequence.

Reconstruction: All images acquired using VD-CAIP + Random-Walk sampling pattern were reconstructed by leveraging temporal information during reconstruction that imposes dynamic priors (sparse and low rank) along the temporal direction with data consistency (Fig 1B), while images acquired using skipped CAIPI were reconstructed using vendor-preinstalled GRAPPA implementation.

Task and Data Processing: For the stimulation, 1 run with 10s blank screen, 15s on /15s off blocks of flashing checkerboard stimuli were acquired using PsychoPy and a 7T-compatible video projector (VPixx Technologies, Inc). Functional analyses were performed with an afni_proc.py script in AFNI. No smoothing was applied and t-statistics were thresholded at p<0.01, spatially and temporal autocorrelation corrected.

Results and Conclusions

Figure 2 shows whole-brain 3D images in the sagittal, coronal, and axial planes acquired with 0.75 and 0.64 isotropic resolutions. The proposed method effectively suppresses image artifacts resulting from temporal leveraging, leading to clear depiction of brain structures compared to skipped-CAIPI. Figure 3 shows the corresponding tSNR maps. The temporal leverage significantly increases the tSNR values by minimizing a trade-off between SNR and accuracy. Figure 4 shows functional activation maps overlaid on the average image. Note that the proposed method shows much higher BOLD sensitivity in the vicinity of gray matter compared to the competing method. We find that 1) a high SNR efficiency across time leads to increased BOLD sensitivity and 2) the segmented 3D EPI coupled with a random sampling scheme can be advantageous for sub-millimeter resolution fMRI smaller than 0.7mm and whole-brain layer fMRI approach.

Acknowledgements

This project is supported in part by the NIH BRAIN Initiative (R01MH111444, U01EB025162), 1R44MH129278 (to Feinberg), and NRF/MSIT No. 2021R1C1C1013603 (to Park).