Resting state MELODIC ICA using fMRIPrep output

Summary of what happened:

I am trying to use output of fMRIprep 23.0.2 with AROMA in FSL’s MELODIC ICA to run a longitudinal resting state analysis (pre/post treatment). I am wondering if I should use this output:

sub-PP_ses-01_task-restingstate_space-MNI152NLin6Asym_desc-smoothAROMAnonaggr_bold.nii.gz

And if adding regressors or using other input is needed?

Moreover, using FSL guide for a multisession concatenated ICA automatically enables co-registration step, so to avoid this I am trying out the bash script below with registration options set to 0.

Lastly, if anyone has experience with multisession ICA in melodic, please advise if you used both sessions or only the first for the group level ICA? My aim is run a group ICA followed by dual-regression on components covering resting-state netowrks using the Smith et al 2009 masks.

Command used (and if a helper script was used, a link to the helper script or the command generated):

OUTPUT_DIR="MELODIC_ICA/output"

# TR(s)
TR=1.56

# Total volumes
NPTS=230

# Number of first-level analyses
MULTIPLE=60

# Input type (cope images from FEAT directories)
INPUT_TYPE=2

# Pre-stats processing
FILTERING_YN=1

# Brain/background threshold, %
BRAIN_THRESH=0

# Motion correction (None)
MC=0

# Slice timing correction (None)
ST=0

# BET brain extraction (None)
BET_YN=0

# Spatial smoothing FWHM (mm)
SMOOTH=0

# Intensity normalization (None)
NORM_YN=0

# Highpass temporal filtering
TEMPHP_YN=1

# MELODIC ICA data exploration
MELODIC_YN=0

# Carry out main stats
STATS_YN=1

# Carry out prewhitening
PREWHITEN_YN=1

# Higher-level modelling (FLAME 1)
MIXED_YN=2

# Higher-level permutations
RANDOMISE_PERMUTATIONS=5000

# Number of EVs
EVS_ORIG=1
EVS_REAL=1
EVS_VOX=0

# Number of contrasts
NCON_ORIG=1
NCON_REAL=1

# Post-stats steps
POSTSTATS_YN=1

# Thresholding (Cluster-based)
THRESH=3
PROB_THRESH=0.05
Z_THRESH=3.1

# Registration settings (Turned Off)
REGINITIAL_HIGHRES_YN=0
REGHIGHRES_YN=0
REGSTANDARD_YN=0

# High pass filter cutoff
PARADIGM_HP=100

# Process each 4D AVW data
INPUT_FILES=(
"/data1/mudarrismaa/Edinburgh_Data/MELODIC_ICA/BOLD_Aroma/sub-PP1_ses-01_task-restingstate_space-MNI152NLin6Asym_desc-smoothAROMAnonaggr_bold"
"/data1/mudarrismaa/Edinburgh_Data/MELODIC_ICA/BOLD_Aroma/sub-PP1_ses-02_task-restingstate_space-MNI152NLin6Asym_desc-smoothAROMAnonaggr_bold"
"/data1/mudarrismaa/Edinburgh_Data/MELODIC_ICA/BOLD_Aroma/sub-PP2_ses-01_task-restingstate_space-MNI152NLin6Asym_desc-smoothAROMAnonaggr_bold"
"/data1/mudarrismaa/Edinburgh_Data/MELODIC_ICA/BOLD_Aroma/sub-PP2_ses-02_task-restingstate_space-MNI152NLin6Asym_desc-smoothAROMAnonaggr_bold"
...
)

# Run FEAT for each input file
for FILE in "${INPUT_FILES[@]}"; do
    echo "Processing: $FILE"
    feat "$FILE"
done

echo "FEAT processing complete."

Version:

fMRIprep 23.0.2; FSL 6.0.4

Environment (Docker, Singularity / Apptainer, custom installation):

Singularity

Data formatted according to a validatable standard? Please provide the output of the validator:

BIDS Validator

To clarify, you want to denoise your fMRI data with AROMA (which involves a MELODIC ICA), and then apply MELODIC? I wouldn’t recommend that. You’ve already run MELODIC within fMRIPrep in order to apply AROMA.

I don’t use MELODIC directly, so I can’t speak to your settings.

Indeed, I want to use melodic AROMA which applies MELODIC ICA to denoise the data, and then use melodic ICA to identify components in my resting state data.

My understanding that these two applications are different and it is possible to use AROMA to denoise the data even through the FSL pipeline without using fMRIprep. Would this be a better/recommended way of doing it?

I don’t think it’s a good idea to run ICA on data that have already been denoised with ICA. Why not just use the ICA components from the ICA-AROMA MELODIC run for whatever you plan to do with the second round of ICA?

I do know that some pipelines perform spatial ICA, followed by temporal ICA on the spatial ICA components concatenated over runs (at least I think that’s how it works), but I think that’s a very different approach from what you’re thinking of.

I believe you can apply AROMA to any preprocessed data, as long as it’s in MNI152NLin6Asym space at 2x2x2 mm resolution.