We didn't worry about this for HCP data with TR=720 msec, though that
doesn't mean someone who knows better than me will agree that was the
right thing to do.
Peace,
Matt.
On 2/28/14 3:29 PM, "sidarta" <[log in to unmask]> wrote:
>Hello,
>
>Our lab is currently in the midst of trying a faster fMRI acquisition
>with a short TR = 500 msec. I'm thinking that with much shorter TR, the
>signal/noise behaviour will change as well, in particular, the way the
>residual autocorrelation should be treated. This is related to whether or
>not my FEAT post-stats outputs are reliable. I'm stuck at this point
>after running the FSL FEAT and see the result. My questions:
>
>1. Can I still use Tukey estimation for autocorr. estimate in film_gls?
>I'm just wondering whether using AR model with higher order may work
>better for shorter TR.
>
>2. In the 2001 paper (Woolrich, et.al.), the Tukey formula M = 2* sqrt(N)
>but after looking at the FSL code, M = 2*sqrt(N)/2 which is just sqrt(N).
>Does this matter much? Sorry I'm not from signal processing background.
>
>Thank you very much!!
>warm regards..
>sidarta
|