Reducing individual differences in task fMRI with OGRE preprocessing for FSL

1330 

Abstract Submission 

Mark McAvoy¹, Lei Liu¹, Ruiwen Zhou¹, Mark McAvoy¹

¹Washington University School of Medicine, St. Louis, MO

Mark McAvoy, PhD  
Washington University School of Medicine
St. Louis, MO

Lei Liu, PhD  
Washington University School of Medicine
St. Louis, MO

Ruiwen Zhou, PhD  
Washington University School of Medicine
St. Louis, MO

Mark McAvoy, PhD  
Washington University School of Medicine
St. Louis, MO

Volumetric preprocessing methods continue to enjoy great popularity in the analysis of functional MRI (fMRI) data. Among these methods, the software packages FSL (Jenkinson et al., 2012) and FreeSurfer (Fischl et al., 2004) are omnipresent throughout the field. However, it remains unknown what advantages an integrated FSL+FreeSurfer preprocessing approach might provide over FSL alone. Here we developed the One-step General Registration and Extraction (OGRE) pipeline to combine FreeSurfer and FSL tools for brain extraction and registration of fMRI data for FSL volumetric analysis.

OGRE preprocessing included coarse registration with FSL tools followed by fine registration with FreeSurfer (Glasser et al., 2013), then formatted for integration with FSL. OGRE preprocessing was compared to traditional FSL preprocessing with a dataset of adult volunteers (N=26) performing a precision drawing task with a MRI-compatible tablet during fMRI scanning. The task followed a blocked design with 15.2s draw/rest, TR = 662 ms, 6×5.4min runs alternating between runs with right hand (RH) drawing and left hand (LH) drawing. Data were preprocessed with FSL or OGRE, and then analyzed with the FSL FEAT general linear model which included explanatory variables for Task (i.e. drawing) and its temporal derivative, each formed by convolving the hemodynamic response with a double-gamma function. Additional variables addressed volumes with excess head motion (framewise displacement > 75th percentile + 1.5* interquartile range). The contrast, Task > rest, was averaged across runs for each participant, and then an across-participants mixed effects analysis was performed to identify a whole-brain voxel-wise effect of Method (OGRE vs. FSL-only) with cluster threshold Z > 3.1 (p < 0.05 corrected). In addition, participant-level data were quantified via a region of interest (ROI) analysis. 42 ROIs were defined from: primary motor representations of hand, lip and foot (Smith & Frey, 2011); the Yeo-7 atlas (Yeo et al., 2011); and the whole brain. For direct statistical comparison of mean signal magnitude and inter-individual standard deviation, these ROIs were entered into a 2 (Method: OGRE vs. FSL-only) * 10 (Area) * 2 (Hemisphere: left, right) * 2 (Condition: LH-draw, RH-draw) repeated measures ANOVA with participants as the random factor.

OGRE preprocessing, compared to traditional FSL preprocessing, led to consistently lower inter-individual variability of task-related BOLD activation (Task > Rest) with a decrease of 44 ± 12%, as shown in Figure 1. This decrease in inter-individual variability was found in all 42 ROIs, and statistically significant (p < 0.001) in 37/42 ROIs (88%). Signal magnitude did not differ systematically between OGRE and FSL: magnitude increased numerically in 28/42 ROIs (67%), and this difference was significant in only 6/42 ROIs (14%). Nevertheless, in a whole-brain analysis, OGRE showed significantly more activation than FSL preprocessing in 5-7 clusters including sensorimotor areas contralateral to movement, while no areas showed more activation for FSL preprocessing than OGRE preprocessing.

The integration of FreeSurfer tools via OGRE preprocessing can improve fMRI data analysis in the context of FSL's volumetric approach. Specifically, OGRE preprocessing led to decreased inter-individual variability, which in turn led to increased detection of task-relevant BOLD activity (sensorimotor areas contralateral to movement). The OGRE pipeline provides a turnkey method to integrate FreeSurfer-based brain extraction and registration with FSL analysis of task fMRI data.

Activation (eg. BOLD task-fMRI) ¹

Methods Development ²

Visuo-Motor Functions

Motor Behavior Other

Workflows

Data analysis

Motor

MRI

Open-Source Software

Other - Task fMRI