Technically, what should matter more is what frequencies dominate the
correlation in the data of interest - if they are lower enough in frequency
than your sample rate (and therefore slice timing spread), then each
latency should produce reasonable maps.  However, this also means that when
you *compare* two different latencies (you might be comparing versus zero
latency), they should generally be separated by more than those same
frequencies (and therefore, by more than the sample rate), or else you are
oversampling them.

A disclaimer, though: I haven't done this kind of analysis, I'm just
working from the principles I know.

Tim


On Tue, Sep 5, 2017 at 8:48 AM, HINDRIKS, RIKKERT <[email protected]>
wrote:

>
> Dear all,
>
> I am analyzing the latency structure of some of the HCP resting-state fMRI
> data and I want to make sure that this makes sense, given that no
> slice-timing correction has been applied to the data. I would appreciate it
> if someone could confirm that the following reasoning is correct:
>
> Since the entire brain is scanned in 0.78 seconds (one sample), it does no
> make sense to analyze signals latencies < 1 sample, because such small
> latencies will be distorted. In particular, it does not make sense to
>  interpolate the cross-covariance functions as done in one of Mitra's
> papers).
>
> However, latencies > 1 sample are distorted only by an amount of < 1
> sample and they can hence be analyzed. So, for example, if two signals have
> a latency of 10 samples, the true latency lies between 9 and 11 samples
> (assumed that the latency is accurately estimated).
>
> Thanks and kind regards,
> Rikkert Hindriks
>
> _______________________________________________
> HCP-Users mailing list
> [email protected]
> http://lists.humanconnectome.org/mailman/listinfo/hcp-users
>

_______________________________________________
HCP-Users mailing list
[email protected]
http://lists.humanconnectome.org/mailman/listinfo/hcp-users

Reply via email to