Dear Cesar
At a TR less than 0.5s the cardiac noise may no longer be broadband. As the cardiac noise is not perfectly sinusoidal there will be a number of higher harmonics each requiring two AR coefficients. Luckily the higher harmonics are mainly present in vessels so with sufficiently small voxels you will mainly have the fundamental in GM voxels, and hence a model order of 5 could be OK, but if you want to be sure you could test this.
Best
Torben
Den 24/12/2009 kl. 18.36 skrev Cesar Caballero:
> Hello,
>
> When modelling the serial correlations with an AR model for the noise in fMRI, it is usually assumed that the maximum model order is about 5 (e.g. Penny et al. NI 2003). However, the same references use acquisition times (TR) values of approximately 2s.
>
> Does anybody know if this value is also valid for very short TRs, e.g. less than 0.5s?
>
> Thanks in advance,
> Cesar
|