Save .tsv file that stores dimensions of .nii.gz files

[JohannesWiesner]

Is your feature request related to a problem? Please describe.

After converting my data to BIDS using bidscoin, some subjects end up with multiple runs for the same modality. This most often occurs for functional scans, where an initial run was aborted and one or more subsequent runs were acquired. In most cases (≈90%), the aborted runs contain an insufficient number of volumes, while the final run contains the expected number.

In practice, this means I can often assume that the last run is the correct one, but this still requires a sanity check. At the moment, I load all .nii.gz files and inspect their shapes using nilearn.image.load_img(path).shape to verify that I only keep runs with the correct number of volumes. However, this approach is quite time-consuming.

Describe the solution you'd like

I am wondering whether this check could be performed on the fly during BIDS conversion in bidscoin. For example, does bidscoin already have access to the number of .IMA files (or slices/volumes) for a given run in memory at some point in the pipeline, such that this information could be stored or exposed? If so, this might allow filtering or flagging incomplete runs without additional computational overhead.

[marcelzwiers]

There is a fix for this namely that you add a requirements to the filesize. It's not perfect but it could work (especially if your run is of a fixed nr of volumes). See e.g. here:

      filesize:                   # File size, e.g. "2[4-6]\d MB" for matching files between 240-269 MB

So what you do is, in your study bidsmap, add a duplicate of the run of interest to your exclude section, and in that duplicate add that filesize constraint. You have to do this with a text editor, or, if you know how (right mouse click), with the bidseditor. Possibly you have to change the provenance field, so that it points to one of your irregular runs. Alternatively, you don't add a duplicate but just add the filesize constraint to your run of interest and then re-run the bidsmapper. If all goes well, then a new run will show up in your study bidsmap, which represent your irregular run (which you can then exclude). But I'mnot sure if this alternative will work (but it's the most easy thing to do)

For permanent change, edit your template bidsmap

[JohannesWiesner]

Ah, nice approach :) But I think I need something more rigorous, as my dataset is quite heterogeneous (scanner parameters can switch between participants, sessions, etc.). So right now I am just doing:

def get_rawdata_df(root):
    layout = BIDSLayout(root,validate=False)
    df = layout.to_df()
    return df

def get_img_dimensions(paths,n_jobs):
    '''Gets the shape of all .nii.gz files. If path does not point to 
    a .nii.gz file just return NaN'''
    
    def _get_shape(path):
        if path.endswith('.nii.gz'):
            return load_img(path).shape
        else:
            return pd.NA
    
    dimensions = Parallel(n_jobs=n_jobs,verbose=10)(delayed(_get_shape)(path) for path in paths)

    return dimensions

df = get_rawdata_df(dir_bidscoined)

# for all .nii.gz files get the file dimensions
df['dimensions'] = get_img_dimensions(df['path'],n_jobs)

which works totatlly fine (just takes very long).