Extract timeseries from all voxels within ROI using Nilearn

Software Support
, ,
1

Extract timeseries from all voxels within ROI using Nilearn

Goal

I am trying to extract the timeseries for all voxels within an ROI using Nilearn. Other topics similar to this seem to be incomplete or the information that was provided was not sufficient enough for me to solve the task given my inexperience with Nilearn.

Intended steps

I want to load in a nifti image

import nibabel as nib

file = "/path/to/file.nii.gz"
data = nib.load(file)
datamat = data.get_fdata()

and then apply NiftiLabelMasker to extract ROIs

from nilearn.masking import NiftiLabelMasker
from nilearn.datasets import fetch_atlas_harvard_oxford

atlas = fetch_atlas_harvard_oxford("cort-maxprob-thr25-2mm")
masker = NiftiLabelsMasker(
    atlas.maps,
    labels=atlas.labels,
    standardize="zscore_sample",
)

masker.fit(file)
masked_data = masker.transform(file)

I then want to use this information to apply it back to the original data. So, if I want the time series for the Superior Parietal Lobule

labels = masker.labels
spl = labels.index('Superior Parietal Lobule')
spl_data = masker.inverse_transform(data[:, np.where(masked_data == spl)]) # something to this effect; this exact line is only meant for conceptual purposes

Another important task I am trying to accomplish is ensuring that this returned data is 2d np.ndarray. So if I call

spl_data.shape
---
(240, 56) # TRs X Voxels

I know that NilearnLabelsMasker.inverse_transform() returns the data into a 3d matrix.

Any help is appreciated.

2

Hi,
If you want to get time courses for all voxels within a given ROI, then you should use a NiftiMasker instance, not the NiftiLabelsMasker, which will create a summary of the signals for all ROIs in the image.
See e.g.

For this you can create a binary mask containing only the ROI you want to consider.
Best,
Bertrand