A new contrast "phase jolt" for analysis of phase-data fMRI reveals strong task-related responses

1922 

Abstract Submission 

Omer Faruk Gulban^1,2, Renzo Huber³, Logan Dowdle⁴, Alessandra Pizzuti¹, Chung Kan³, Rainer Goebel^1,2, Dimo Ivanov¹, Kendrick Kay⁴

¹Maastricht University, Maastricht, Netherlands, ²Brain Innovation, Maastricht, Netherlands, ³National Institutes of Health, Washington, MD, ⁴Center for Magnetic Resonance Research, Department of Radiology, University of Minnesota, Minneapolis, MN

Omer Faruk Gulban  
Maastricht University|Brain Innovation
Maastricht, Netherlands|Maastricht, Netherlands

Renzo Huber  
National Institutes of Health
Washington, MD

Logan Dowdle, Ph.D.  
Center for Magnetic Resonance Research, Department of Radiology, University of Minnesota
Minneapolis, MN

Alessandra Pizzuti  
Maastricht University
Maastricht, Netherlands

Chung Kan  
National Institutes of Health
Washington, MD

Rainer Goebel  
Maastricht University|Brain Innovation
Maastricht, Netherlands|Maastricht, Netherlands

Dimo Ivanov  
Maastricht University
Maastricht, Netherlands

Kendrick Kay  
Center for Magnetic Resonance Research, Department of Radiology, University of Minnesota
Minneapolis, MN

The vast majority of fMRI studies examine only the magnitude component of the fMRI data while discarding the phase component. Although phase data are known to be sensitive to task activation [1-4], their practical usefulness is impeded by processing difficulty compared to magnitude data. For instance, the circular nature of phase spanning 2π radians range is not handled properly in most preprocessing and analysis programs. Here, we introduce using the second spatial derivative of phase images ("phase jolt") instead of the phase images themselves as the basis for functional time series analyses. Phase jolt computation casts circular 2π range values of phase with no natural 0 point onto a new range of values with 0 to π range, where noise is centered around π/2. This allows straightforward application of preprocessing and analysis methods developed for magnitude time series on phase jolt time series. We show that there is surprisingly large signal in phase jolt time series corresponding to activation sites in task fMRI experiments.

We used 6 different 7 T fMRI datasets, all of which included both magnitude and phase data:

- Dataset 1: Visual stimulation adapted from github.com/VPNL/fLoc, with data acquisition similar to [5]. Each stimulus lasted 4 s. Functional images were collected with 1.8×1.8×1.8 mm3 voxels at TR=1.6 s using gradient echo EPI (2D). This dataset is provided by K.K.
- Dataset 2: Visual stimulation using 4 center-surround probes located in the upper/lower and left/right visual fields. Each stimulus lasted 12 s followed by 8.4 s of rest. Functional images are collected with 1×1×1 mm3 voxels at TR=850 ms using gradient echo EPI (2D). This dataset is provided by L.T.D. (RF1 MH117015 grant).
- Dataset 3, 4: Visual stimulation with faces, objects, and their scrambled versions. Each stimulus lasted 10 s. Functional images were collected with 1.6×1.6×1.6 mm3 or 0.8×0.8×0.8 mm3 voxels at TR=650 ms or TR=1850 ms using gradient echo EPI (2D) in experiments 3 and 4, respectively. This dataset is provided by L.T.D. (RF1 MH117015 grant).
- Dataset 5: Visual stimulation with flashing checkerboards. Each stimulus lasted 32.7 s with equal amount of rest afterwards. Functional images are collected with 0.8×0.8×0.8 mm3 voxels at TR=8.1 s using T1234 (3D EPI) [6]. This dataset is provided by R.H & C.K.
- Dataset 6: Finger tapping stimulation. Each stimulation lasted for 33 s with equal amounts of rest afterwards. Functional images are collected with 0.75×0.75×1.25 mm3 voxels at TR(BOLD)=3.3 s using SS-SI VASO (3D EPI). This dataset is from [7].

Magnitude fMRI time series were only motion corrected and high pass filtered (in BrainVoyager v23 [8]). Phase fMRI time series were first subjected to phase jolt computation newly implemented in the LayNii v2.6 [9] LN2_PHASE_JOLT program (with '-2D' option for datasets 1-4, and 6). Afterwards, the phase jolt fMRI time series were preprocessed in the same way as the magnitude time series. General linear model was performed to compute the activity maps (stimulation vs baseline).

See Fig1-2 captions.

We have shown that through the simple calculation of phase jolt, phase data can be easily analyzed to reveal substantial task activations. These activations appear to be complementary to standard magnitude time series activations. Phase jolt activity tends to be located near and outside of the cortical gray matter, presumably indicating sensitivity to pial veins. This might be leveraged to detect and mitigate the venous contributions from BOLD magnitude. Alternatively, the strong task-related activity in phase jolt time series might be valuable as an additional information for brain decoding analyses [10]. In the future we will focus on characterizing the biophysical mechanisms underlying phase jolt with respect to MR signal acquisition and noise properties. With further development, we suggest that the effectiveness of phase jolt analysis may spur the usage of phase data in fMRI studies.

Methods Development ¹

BOLD fMRI ²

Imaging Methods Other

Blood

Cortex

Data analysis

fMRI CONTRAST MECHANISMS

FUNCTIONAL MRI

HIGH FIELD MR

MRI

MRI PHYSICS

Vision

Other - Phase