VBM by ROI in nilearn

1

Hi all, Is it possible with nilearn to obtain volume values by ROIs from a T1 that has been preprocessed on fmriprep?
Best regard,
Raul

2

Hi,
Yes, it is quite possible, using a NiftiMasker, providing as mask the binary image the defines the ROI. Please take time to read

HTH,
Bertrand

1 Like
3

Ahoi hoi @rglezgz,

just to follow up on @bthirion’s comment, depending on your data and the planned analyses, different options of nilearn.maskers could be feasible/interesting. For example, if you want to extract volume values from multiple ROIs and participants, the following example could be helpful:

# import modules

from nilearn import datasets
from nilearn.input_data import NiftiLabelsMasker
import nibabel as nb

# download VBM data from 10 example participants 
vbm_exp = datasets.fetch_oasis_vbm(n_subjects=10)

# download an example atlas, here AAL
aal_atlas = datasets.fetch_atlas_aal()

# set up NiftiLabelsMasker to extract volume values from multiple ROIs (the AAL atlas)
aal_masker = NiftiLabelsMasker(labels_img=aal_atlas.maps)

# define an empty list to store volume values for each participant
vbm_exp_data = []

# loop over participants, respectively extract and storing volume values
for i,vbm_img in enumerate(vbm_exp['gray_matter_maps']):
    
    # load the current image using nibabel
    vbm_exp_img = nb.load(vbm_img)
    
    # extract volume value per ROI from current image and append it to the list
    vbm_exp_data.append(aal_masker.fit_transform(vbm_exp_img).squeeze())

# convert the list into a pandas DataFrame, assigning atlas labels as columns
df_vbm_data = pd.DataFrame(vbm_exp_data, columns=aal_atlas.labels)

With this relatively straightforward approach, you get a DataFrame that can be utilized for statistical analyses and/or visualization. For example, showing volume values for all participants per ROI:

# import modules

import seaborn as sns

# set figure parameters
sns.set(rc={"figure.figsize":(22, 4)}) 
sns.set_style('whitegrid')

# create plot
g = sns.stripplot(df_vbm_data)

# set some aesthetics 
sns.despine(offset=10)

# rotate labels for readability
g.set_xticklabels(labels=df_vbm_data.columns, rotation=90);

# assign axis labels
g.set_ylabel('Volume');
g.set_xlabel('ROI');

image
image1291×378 93.7 KB

HTH, cheers, Peer

4

Excellent, thanks to both. @PeerHerholz that script is very helpful, but I still need to obtain the TIV values to adjust the ROI volumes.

5

TIV can be found in the FreeSurfer outputs /path/to/Freesurfer/sub-XX/stats/brainvol.stats or aseg.stats.

6

Hi @Steven, I don’t understand how to find the TIV value. I just ran the script of @PeerHerholz and I can’t find these outputs.
I want to use this script on a dataset preprocessed by fmriprep, but for T1 there is no TIV as an output.

thanks for you time

7

Did you run FreeSurfer as part of fmriprep? If so, you can find the TIV in the statistics text file output by FreeSurfer in the path I described in my previous post. If not, you can run recon-all from FreeSurfer on your T1, which will output the TIV (and other stats) in the text files I mentioned in my previous answer.

8

Thanks, @Steven. I just used --fs-no-reconall to expend less time in the fmriprep analysis. I’m running now.