You need to apply 3dmaskave with the mask being defined by the ROI dataset and the input dataset being a timeseries dataset defined on the same 3D grid. For example:
3dmaskave -mask roi+orig -q epi+orig > aver.1D
For each sub-brick in the input dataset (epi+orig), the values from voxels which are nonzero in the mask (roi+orig) will be averaged and sent to stdout -- you can capture these to a file with the '> aver.1D' redirection. You can then plot that file with something like
1dplot aver.1D
If your roi+orig dataset has multiple ROIs defined in it, each with a different voxel value, then you can select a subset of the nonzero voxels with the '-mrange' option in 3dmaskave. As usual, see the output of
3dmaskave -help
for all the details.