Hi, Zixun-

OK, thanks for sharing the data.

A couple things to note: The EPI dataset has a wide range of values, from -637911 to 314724, but it also includes supertiny, nonzero values, of order 10**-28 and 10**-31. The mask that is being used with it also doesn't seem to perfectly fit the data.

The large range of values isn't really a problem, but I am surprised to see it. The source of the calfun errors is having entire time series that have supertiny nonzero values---the mask doesn't remove those. Also, automasking didn't work well with this dset because it has a somewhat filamentary spatial structure. So, to generate a mask where time series were non-zero and non-supertiny, I did the following:

3dcalc                                                                       \
    -a          DSET_EPI'[0]'                                                \
    -expr       'step(a-0.001)+step(-a-0.001)'                               \
    -prefix     mask_estimate2.nii.gz                                        \
    -overwrite

(Note that you could adjust the windowing around zero, which is masking out anything that has magnitude <0.001.)

After doing that, I reran your 3dFWHMx command:

3dFWHMx                                                                      \
    -detrend  2                                                              \
    -mask  mask_estimate2.nii.gz \
    -dset     sub-012_task-motor_acq-ME5_negDelta_R2_star.nii.gz             \
    -acf      sub-012_task-motor_negDelta_R2_star    >> output3dFWHMx

... and there were no calfun errors. Therefore, those were a result of the supertiny nonzero values, or quasi-zero time series, being present in the non-masked region.

The resulting ACF image looked pretty reasonable, I would say. Though, I will note again the caveat that this program was designed for working on residual datasets.

--pt