Hi Robin,

Sorry for getting a bit distracted.

I was not expecting you to run 3dDeconvolve on a 4D dataset, ever, but to do something like:

- extract an ROI average time series of choice
(the realtime plugin can send this (or multiple)) to an IP address

- regress out the motion parameters (and polort) on the full time series
(run 3dDeconvolve on the 1D average)

- perform your comparison computation on the windowed time series


Having 429 time points might be managable.

Also, note that projecting out a few time series (motion, polort, sinusoids) from a 1D file might be faster to do directly in python (scipi.linalg). While a compiled C program is generally faster, you have to deal with writing to and reading from the file system, as well as repeatedly starting up the external program (3dDeconvolve). This overhead might make it more reasonable to do via scipy.linalg, since it is just a simple projection of unwanted terms from one time series.

- rick