Removing Motor Response Artifacts in fMRI Experiments

In my fMRI experiment, participants were instructed to perform motor responses for a task during the interval between blocks. These responses, however, were not of interest as the task solely aimed at improving concentration levels. Consequently, no timestamps were recorded for these motor responses.
However, in our group analysis, we observed significant effects of these motor responses: all experimental conditions exhibited negative activations in regions of interest.
This could potentially be attributed to motor responses between blocks, leading to a higher baseline in the GLM.
Hence, I’m wondering if there are any noise-reduction techniques available to eliminate such noise from the data?

Dear Sata,
You can try and use nilearn.image.high_variance_confounds - Nilearn
It estimates high-variance signals present in the data that could be removed.