NiftiLabelsMasker extract timeseries for talairach atlas ROIs

I’m trying to extract timeseries for ROIs defined by Talairach atlas from nilearn.
I define a masker object

talairach = datasets.fetch_atlas_talairach('ba')
talairach_filename = talairach.maps
talairach_masker = NiftiLabelsMasker(labels_img=talairach_filename, standardize=True, 
                                     resampling_target = 'data',
                                     memory='nilearn_cache', verbose=5)

Then extract timeseries using fit_transform

timeseries = talairach_masker.fit_transform(denoised_data_path, confounds=None)

The resulting variable timeseries has shape (n_scans, n_rois). In my case (187, 69).
But len(talairach.labels) is 72, with the first label being background. So timeseries should be (187, 71)

If I specify resampling_target = 'lables', it seems to work correctly, and the output is (187, 71).

I tried the same with Harvard-Oxford atlas. And there it doesn’t matter how I specify the resampling_taget, the output always has the correct shape.

Does anybody know why it behaves differently?

I believe when resampling the talairach atlas to your image some talairach regions end up being empty (containing 0 voxels), and in that case they would not be allocated an output dimension by the masker