Indeed, that seems totally normal.
At 13 time points, you have half as many regressors as time point. The linear regression will in general look very noisy at the beginning (at least one of the motion regressors will probably fit that gentle rise that you are looking for), and will start to stabilize after sufficient time to have useful variance in the more "ideal" time series you are looking for. In this case, I would expect you to start getting reasonable results around time point 50, when the signal of interest has had initial time to rise and fall.
We have a couple of sample datasets that you can review, the older of which (in AFNI_data6) shows a similar phenomenon to yours.
If you get the
AFNI_data6 package, run the real-time demo #2 under:
AFNI_data6/realtime.demos/demo_2_feedback
There is a README file there.
For a different real-time demo package, consider running:
@Install_APMULTI_Demo2_realtime
But that package does not have any real-time regression examples, not yet. It also does not have actual multi-echo examples, though the data is there and it would take only a tiny effort to code an example (more time to set up regression testing (not linear regression), which is part of the purpose of the demo).
- rick
