Without the data, it is difficult to say for sure. Guesses:
- 3dfim+ is very very old -- I don't know if it is still "good". Try 3dTcorr1D instead.
- Is it possible ANTS left a "hole" in the data when transforming it (that is, created a voxel with all zero entries)? It would be necessary to look at the transformed time series dataset (input to 3dfim+) at a location where the correlation was 0 to investigate the answer to this question.
Without more information, I don't know what else to say.