3296
Feasibility of whole-brain functional MRI with 1-mm isotropic resolution at 3T
Wei-Tang Chang1,2,3, Stephanie Langella4, Min Sung Seo5, Weili Lin2, and Kelly Giovanello5
1Biomedical Engineering, UNC at Chapel Hill, Chapel Hill, NC, United States, 2Biomedical Research Imaging Center, UNC at Chapel Hill, Chapel Hill, NC, United States, 3Radiology, UNC at Chapel Hill, Chapel Hill, NC, United States, 4Psychiatry, Harvard Medical School, Boston, MA, United States, 5Psychology & Neuroscience, UNC at Chapel Hill, Chapel Hill, NC, United States
1Biomedical Engineering, UNC at Chapel Hill, Chapel Hill, NC, United States, 2Biomedical Research Imaging Center, UNC at Chapel Hill, Chapel Hill, NC, United States, 3Radiology, UNC at Chapel Hill, Chapel Hill, NC, United States, 4Psychiatry, Harvard Medical School, Boston, MA, United States, 5Psychology & Neuroscience, UNC at Chapel Hill, Chapel Hill, NC, United States
Synopsis
Keywords: Data Acquisition, Neuro
fMRI with ultrahigh resolution has typically been acquired at 7T in order to overcome the inherently low SNR at 3T. However, 7T scanners are inaccessible to the majority of investigators. In this study, we developed new acquisition and reconstruction approaches and employed the denoising method NORDIC to improve the resolution of 3T fMRI to 1 mm isotropic with whole-brain coverage and volumetric scan time of 2 seconds. Our results showed improved tSNR and low spatial blurring. The detection capability of resting-state networks was comparable to that of 7T. BOLD activations in cortical and subcortical regions were observed in task-based fMRI.Introduction
Whereas the 2-mm resolution in HCP protocol demonstrated the successes in precise functional mapping1, fMRI with ultrahigh resolution (≤ 1 mm) for investigation of fine-scale structures is of great interest and still under pursuit. To date, most of the ultrahigh-resolution fMRI acquisitions were acquired at 7T for higher signal-to-noise ratio (SNR). However, 7T scanners are inaccessible to the majority of investigators. In this study, we 1) improved the spatial resolution of 3T fMRI to 1 mm isotropic with whole-brain coverage and volumetric scan time of 2 seconds, and 2) examined the feasibility comprehensively. Specifically, we developed new approaches for acquisitions and reconstructions to improve the acceleration capability and employed the NORDIC denoising method2 to improve SNR. The detection sensitivity of resting-state networks at 3T was compared with the sensitivity at 7T. Finally, we evaluated BOLD activations in cortical and subcortical regions during episodic memory tasks.Method
In conventional SMS acquisitions with both inplane and slice accelerations, the accelerated image was reconstructed in two sequential steps as shown in Figure 1a. The reconstruction error in the first reconstruction step may propagate to, and be enhanced by, the next step. In this study, we combined the two reference scans for slice and inplane reconstructions into one. This allowed the accelerated images to be reconstructed in one step as shown in Figure 1b, thereby lowering the g-factor penalty and increasing the image fidelity. The kx lines skipped by partial-Fourier acquisition was reconstructed using the projection-onto-convex-set (POCS) method. The spatial blurring was measured by the full-width-half-magnitude (FWHM) of the noise kernel3.Thirty-four healthy adults with ages ranging from 18 to 80 years were recruited for 3T experiments. Four datasets were excluded due to serious head movement. MR images were acquired using a 3T Prisma scanner (Siemens Healthcare, Erlangen, Germany) with the following parameters: SMS factor = 6, in-plane acceleration rate = 2, TR = 2 seconds, FOV = 176×176×126 (R-L×H-F×A-P) mm3, number of axial slices = 126, TE = 32 ms. One 10-minute resting-state scan and six 5.4-minute task-based scans were acquired. The in-scanner task was a continuous recognition version of the Mnemonic Similarity Task4. To compare with the publicly-available multiband fMRI sequence distributed by Center for Magnetic Resonance Research (CMRR), we also collected the fMRI with similar parameters above except: FOV = 192×192×126 (R-L×H-F×A-P) mm3.Results
Figure 4a and the left panel of Figure 4b showed that the joint GRAPPA demonstrated fewer artifacts and higher signal intensity around the subcortical regions than the two-step GRAPPA reconstruction. After NORDIC denoising and POCS reconstruction, the image became less grainy and had reduced ringing artifacts as shown in the middle and right panels of Figure 4b.For the group-level quantitative analysis, we examined tSNR and spatial blurring as shown in Figure 3. After NORDIC denoising, the tSNR of 3T was enhanced from 7.5 to 36.6 and significantly higher than that of 7T (p < 0.001). The results in subcortical grey matter were likewise. Meanwhile, the averaged point spread functions across dimensions remained as ~10% and showed no significant difference compared to 7T (p = 0.17). Our results indicated that the enhancement of tSNR by NORDIC did not compromise the spatial resolution.
Next, we examined the detection capability of resting state networks at 3T. For a single individual, Figure 4a demonstrated the detection capability of resting-state networks by ICA. For group level (N=7), the seed-based FC analysis was performed on both 3T and 7T datasets for comparison as shown in Figure 4b. The detection capability of DMN was further characterized using receiver operating characteristic (ROC) and quantified by the area under ROC curve (AUC), as shown in Figure 4c. The AUC of 3T with NORDIC showed only 7.4% reduction from that of 7T. Our results suggested the 1-mm resolution fMRI at 3T was able to detect RSNs even with a small number of participants.
For task-based fMRI, we employed an object recognition memory task to elicit cortical and subcortical activations. The results in Figure 5a demonstrate the typical BOLD activations as reported in prior episodic memory studies5–7. Figure 5b demonstrates the subcortical activations which was previously reported to be associated with memory retrieval8,9. Our results of evoked BOLD responses suggested that the 1-mm fMRI at 3T was able to detect the BOLD activations in both cortical and subcortical regions.
Discussion
This study improved the spatial resolution of fMRI at 3T to 1 mm isotropic with whole-brain coverage and TR of 2s. Compared with the publicly-available multiband sequence, the quality of image reconstruction was substantially improved by the newly-designed acquisition and reconstruction methods. Our results demonstrated significantly improved tSNR with low spatial blurring. The resting-state fMRI at 3T demonstrated typical RSNs in both individual and group levels. The detection capability of RSNs at 3T, measured by AUC, was reduced by only 7.4% compared to that of 7T without NORDIC. In the task-based fMRI, our 3T data demonstrated the BOLD sensitivity in both cortical and subcortical regions with reasonable sample size. With the demonstrated sensitivity and low spatial blurring, our imaging methods on 3T holds great promises in providing better imaging features for the connectomic studies involving fine-scale subcortical structures.Acknowledgements
This work was supported in part by NIH grant R21AG060324.
References
1. Van Essen, D. C. et al. The WU-Minn Human Connectome Project: An overview. NeuroImage 80, 62–79 (2013).
2. Vizioli, L. et al. Lowering the thermal noise barrier in functional brain mapping with magnetic resonance imaging. Nat. Commun. 12, 5181 (2021).
3. Huber, L. R. et al. LayNii: A software suite for layer-fMRI. Neuroimage 237, 118091 (2021).
4. Stark, S. M., Kirwan, C. B. & Stark, C. E. L. Mnemonic Similarity Task: A Tool for Assessing Hippocampal Integrity. Trends Cogn. Sci. 23, 938–951 (2019).
5. Grady, C. L., McIntosh, A. R. & Craik, F. I. M. Age-related differences in the functional connectivity of the hippocampus during memory encoding. Hippocampus 13, 572–586 (2003).
6. Uncapher, M. R. & Wagner, A. D. Posterior parietal cortex and episodic encoding: Insights from fMRI subsequent memory effects and dual-attention theory. Neurobiol. Learn. Mem. 91, 139–154 (2009).
7. Wing, E. A., Ritchey, M. & Cabeza, R. Reinstatement of Individual Past Events Revealed by the Similarity of Distributed Activation Patterns during Encoding and Retrieval. J. Cogn. Neurosci. 27, 679–691 (2015).
8. Giovanello, K. S. & Dew, I. T. Z. Relational Memory and its Relevance to Aging. in The Wiley Handbook on the Cognitive Neuroscience of Memory (eds. Addis, D. R., Barense, M. & Duarte, A.) 371–392 (John Wiley & Sons, Ltd, 2015). doi:10.1002/9781118332634.ch18.
9. Kim, H. Neural activity that predicts subsequent memory and forgetting: A meta-analysis of 74 fMRI studies. NeuroImage 54, 2446–2461 (2011).
Figures
Figure 1: The pulse sequences of reference scans and the corresponding reconstruction approaches. (a) Separated reference scans and image reconstructions for slice and inplane reconstructions. (b) The combined reference scan and joint image reconstruction. The solid and hollow circles represent the existed and skipped kx-lines respectively. Abbreviations: Gs – slice gradient; Gp – phase gradient; Gr – readout gradient; Acc – accelerated; Ref – reference; recon – reconstructed.
Figure 2: Reconstructed images by (a) two-step GRAPPA and (b) joint GRAPPA were shown in the left two panels. The reconstructed image of joint GRAPPA was then denoised and partial-Fourier reconstructed as shown in the 3rd and 4th panel from the left.
Figure 3: The group-level tSNR and point-spread function after partial-Fourier reconstruction (N=7). (a) The tSNRs of 3T, 7T and 3T with NORDIC denoising in cortical and subcortical grey matters. The cortical and subcortical regions are highlighted by the red color in the brain. (b) The point spread function is measured by the ∆FWHM. The ∆FWHM of 7T and 3T with NORDIC were compared. The dashed line highlighted the 10% of point spreading. Abbreviations: FWHM – full-width half-magnitude; FE – frequency encoding; PE – phase encoding; SLC – slice; Avg. – averaged.
Figure 4: The detection capability of resting-state networks at individual and group levels. (a) The ICA-based RSNs identified from one representative individual. The length of 3T fMRI time series was 10 minutes. (b) The seed-based DMN with the seed placed at PCC as indicated by the solid blue circle. Two different groups of 7 participants were recruited for 3T and 7T experiments. (c) The AUCs for 3T and 7T were noted in the plot. Abbreviations: MVN – medial visual network; LVN – lateral visual network; TPR – true positive rate; FPR – false positive rate.
Figure 5: The group-level statistical maps of BOLD activations evoked by Mnemonic Similarity Task (N=30). (a) The statistical maps of the cortical BOLD activation associated with all the trials of correct responses. (b) The BOLD activations in hippocampus. The z scores of BOLD responses were encoded as indicated by the color bar. All the statistical maps were corrected by multiple comparison (corrected p < 0.05). The zoom-in views in (b) were magnified from the red blocks. The green arrows highlight the hippocampal activations.
DOI: https://doi.org/10.58530/2023/3296