Differentiating neural and non-neural components in fMRI using cross-cortical depth delay patterns

Poster No:

Th583 

Submission Type:

Abstract Submission 

Authors:

Jingyuan Chen1, Anna Blazejewska1, Nina Fultz2, Bruce Rosen3, Laura Lewis4, Jonathan Polimeni1

Institutions:

1A. A. Martinos Center for Biomedical Imaging, Harvard Medical School, Massachusetts General Hospital, Charlestown, Boston, MA, 2A. A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, Boston, MA, 3Department of Radiology, A.A. Martinos Center for Biomedical Imaging, MGH and Harvard Medical School, Charlestown, Boston, MA, 4Boston University, Boston, United States

Introduction:

As neural signals initiate within the parenchyma, and the elicited BOLD signal propagates downstream along draining veins toward the pial surface [1,2], BOLD-fMRI time courses from superficial cortical layers should lag those in deeper layers. In contrast, such a temporal progression may not exist in BOLD fluctuations driven by non-neuronal effects. Therefore, we hypothesize that cross-cortical-depth temporal lag patterns may inform whether a specific fluctuation is neuronal in origin, and demonstrate their promise in de-nosing medium- and high-resolution task data. Additionally, we show that restricting analyses to deep cortex can mitigate systemic physiological fluctuations, and may potentially improve the spatial specificity of fMRI results.

Methods:

Data: 8 subjects participated in this study, each underwent 3 visual task scans (8 blocks, 15/20s on/off) and 3 resting state (RS) scans (12~15 min) with graded spatial/temporal resolutions: voxel size=1.1/1.5/2.0 mm iso., TR=1.7/1.12/0.93 s, FA=72/62/58o, TE=26 ms, SMS factor=3 (7T, Siemens Healthineers), with a custom-built 32-channel coil.

Analysis I: Surface-based cortical depth estimation was performed with a 0.75 mm iso. FOCI-MEMPRAGE data [3] using FreeSurfer, following which the normalized cortical depth ('0':white matter; '1':pial) of each EPI voxel was computed according to its centroid coordinates [4]. After rigid-body realignment, each scan was spatially decomposed using MELODIC ICA. Within each IC, voxels were separated into five groups based on their normalized depths (D1:0–40%, D2:40–80%, D3:80–120%, D4:120–160%, D5:160–200%, where depths >100% are above the pial surface). Relative delays across the mean signals of each depth were then estimated using temporal cross-correlation. Finally, to characterize whether superficial depths lag deeper cortex, two metrics were employed: (NF-M1) the relative delay vector from depths D1–D5 was correlated with an order vector (1:5) using Spearman's correlation coefficient r, and ICs with low values of r (r<0.2) were considered non-neural; (NF-M2) ICs with the relative D1-D5 lag shorter than 0.2 s were considered non-neural. Noise ICs identified by both metrics were included in GLM to infer task activation using SPM's FAST [5].

Analysis II: Fractional contributions from systemic physiological noise (modelled by RETROICOR [6] and RVHRCOR [7]) across cortical depths were quantified for the RS scans. RS networks (RSNs) were resolved using ICA, then projected onto the cortical surface at different depths for visualization.

Results:

I: Fig. 1A illustrates the cross-cortical-depth delay patterns of neural and non-neural ICs. Such distinct patterns are salient in each individual's IC results, and can be used to isolate signal and noise (Fig. 1B). After accounting for these noise ICs, a notable trend of enhanced task activation is achieved (Fig. 1C 'Raw' vs. 'NF-M1/2'), suggesting the efficacy of the proposed de-noising approach.

II: Physiological noise is most pronounced and spatially extensive near the pial surface (Fig. 2A). Key regions within each RSN are found in all cortical depths, with patterns measured from within the parenchyma being more focal than those measured at the pial surface (Fig. 2B).

Conclusions:

We show that cross-cortical-depth lag patterns show promise for automatically identifying neural and non-neural fluctuations in medium- or high-resolution data that cortical depth information could be resolved. It is roughly analogous to multi-echo (ME) ICA [8] for separating neural and non-neural fluctuations, and may serve as an alternative in scenarios in which ME acquisitions are not preferable: e.g., high spatiotemporal resolutions, high field strength (due to ultra-short tissue TEs) and non-BOLD contrasts (CBV). Our preliminary characterization of physiological noise and RSNs across cortical depths suggest that excluding voxels intersecting the pial surface can reduce physiological effects and improve neuronal/spatial specificity.

Imaging Methods:

BOLD fMRI

Modeling and Analysis Methods:

Exploratory Modeling and Artifact Removal 2
Methods Development
Motion Correction and Preprocessing 1
Task-Independent and Resting-State Analysis

Keywords:

Cortical Layers
Data analysis
FUNCTIONAL MRI
HIGH FIELD MR
Machine Learning
Modeling
Multivariate

1|2Indicates the priority used for review
Supporting Image: Figure1.png
Supporting Image: Figure2.png
 

My abstract is being submitted as a Software Demonstration.

No

Please indicate below if your study was a "resting state" or "task-activation” study.

Resting state
Task-activation

Healthy subjects only or patients (note that patient studies may also involve healthy subjects):

Healthy subjects

Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.

Yes

Was any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.

Yes

Was any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.

Not applicable

Please indicate which methods were used in your research:

Functional MRI

For human MRI, what field strength scanner do you use?

7T

Which processing packages did you use for your study?

AFNI
SPM
FSL

Provide references using author date format

[1] Yu X. (2012). Direct imaging of macrovascular and microvascular contributions to BOLD fMRI in layers IV-V of the rat whisker-barrel cortex. Neuroimage. 59(2):1451-1460.
[2] Lewis L.D. (2018). Stimulus-dependent hemodynamic response timing across the human subcortical-cortical visual pathway identified through high spatiotemporal resolution 7T fMRI. Neuroimage. 181:279-291.
[3] Zaretskaya N. (2018). Advantages of cortical surface reconstruction using submillimeter 7 T MEMPRAGE. Neuroimage. 165:11-26.
[4] Polimeni, J. R. (2018). Analysis strategies for high-resolution UHF-fMRI data. NeuroImage, 168, 296-320.
[5] Corbin N. (2018). Accurate modeling of temporal correlations in rapidly sampled fMRI time series. Hum Brain Mapp.
[6] Glover G. H. (2000). Image‐based method for retrospective correction of physiological motion effects in fMRI: RETROICOR. Magnetic Resonance in Medicine, 44(1), 162-167.
[7] Chang C. (2009). Influence of heart rate on the BOLD signal: the cardiac response function. Neuroimage, 44(3), 857-869.
[8] Kundu P. (2011). Differentiating BOLD and non-BOLD signals in fMRI time series using multi-echo EPI. Neuroimage 60(3):1759-70.