Containerized pipeline for handling multi-echo fMRI data.

2267 

Abstract Submission 

Andre Zugman¹, Dylan Nielson², Safa Rahman¹, Marie Zelenina¹, Charles Lynch³, Conor Liston⁴, Daniel Pine¹

¹National Institute of Mental Health, Bethesda, MD, ²National Institute of Mental Health, Washington, DC, ³Weill Cornell Medical College, New York, NY, ⁴Weill Cornell Medicine, Cornell University, New York, NY

Andre Zugman  
National Institute of Mental Health
Bethesda, MD

Dylan Nielson, PhD  
National Institute of Mental Health
Washington, DC

Safa Rahman  
National Institute of Mental Health
Bethesda, MD

Marie Zelenina  
National Institute of Mental Health
Bethesda, MD

Charles Lynch  
Weill Cornell Medical College
New York, NY

Conor Liston  
Weill Cornell Medicine, Cornell University
New York, NY

Daniel Pine  
National Institute of Mental Health
Bethesda, MD

Neuroimaging researchers have available a number of software tools and preprocessing options. Recently docker and singularity have facilitated the use and sharing of containers dedicated to processing neuroimaging data, most noticeably fMRIprep has been widely adopted as it provides a robust pipeline that can be evoked in a single command line argument (Esteban et al., 2019). However, there might be particular cases when it is desirable to have alternate pipelines available to researchers. Here we present a docker container that implements the pipeline described in (Lynch et al., 2020). This container aims to fill an existing gap. Although there are many processing software and pipelines available, there is no dedicated container for end-to-end processing and denoising of multi-echo fMRI. This pipeline previously showed good reliability with multi-echo fMRI data projected to the cortical surface.

The functional processing requires the anatomical processing of the HCP pipeline (Glasser et al., 2013). In short it consists of a pre-FreeSurfer step, which includes the alignment of the T1w and T2w scans, bias field correction and registration from subject to MNI space. This is followed by FreeSurfer processing, and lastly the final anatomical format in GIFTI or NIFTI are produced. The container allows for the researcher to run these steps, or to run only the functional processing if HCP style outputs are already available.
The fMRI pipeline consist of i) preprocessing fieldmaps (averaging available fieldmap, TOPUP, aligning to subject space and brain extraction) ; ii) slice timing and head motion correction of the functional images; iii) signal decay based denoising and removal of spatially diffuse noise, using tedana. (DuPre et al., 2021; Kundu et al., 2012); iv) and mapping the denoised signal to surface space.

We present a working docker container that is ready to run the aforementioned pipeline. The container entrypoint takes the pipeline (anatomical or functional), participant folder , path to the data (as mounted in the container), and number of cores to use as arguments. Currently the command line requires that the user specifies the correct mount point, FreeSurfer and MATLAB licenses.
An example command is:
"""
Docker run --name test \
-v /PATH/TO/DATA/::/data \
-v /PATH/TO/OUTPUT/:/out \
-v /PATH/TO/WORK:/work \
-v ~/Documents/licenses/license.txt:/opt/freesurfer-6.0.1/license.txt \
-e MLM_LICENSE_FILE=9999@your.license.server \
--platform linux/amd64 listonpipeline <-anat/-func> -p [participant] -d [/data/] -c [number of cores]
"""
This container was tested on data now available online on openneuro (https://openneuro.org/datasets/ds004787/versions/1.1.0.). This data consists of 5 healthy volunteers, scanned in a GE MR750 3T scanner. Multi-band Multi-echo resting-state fMRI was acquired with the following parameters (2.5mm isotropic, TR=2.5 s; TEs= (12.9 ms, 32.2 ms, 51.6 ms, 70.9 ms); Flip Angle= 77 degrees; in-plane acceleration=3, multi-plane acceleration=2). High-resolution T1w and T2w scans were collected in the same session. All subjects gave informed consent and consented to data sharing (protocol 01-M-0192).

We provide a working container for preprocessing multi-echo fMRI and outputting fMRI signals in surface space. We intended to improve this container by removing the MATLAB license requirement and adding BIDS compatibility. The current version uses the automatic classification of noise components, tedana also allows for the manual classification of the components. Future versions should allow researchers to use manually selected components and re-run only the relevant steps of the pipeline.

Workflows ¹

Informatics Other ²

Data analysis

Workflows

Other - Container