I don't know if this is something you're interested in, but it would be awesome if pySPFM didn't require AFNI to run. I could be wrong, but it looks like the only places where AFNI is called are in hrf_afni and update_header:

https://github.com/eurunuela/pySPFM/blob/d69907cd3a8f0455b44504c74a73ee93915d4409/pySPFM/deconvolution/hrf_generator.py#L72

https://github.com/eurunuela/pySPFM/blob/dca806578adc86f55943a708d18bf02a2d4155ab/pySPFM/io.py#L62

It seems like hrf_afni could be replaced with something from nilearn.glm (e.g., nilearn.glm.first_level.spm_hrf). I don't know about update_header, but updating a HEAD/BRIK pair's header seems like something nibabel should be able to do.