Hi Catherine -
Sorry for the slow response
On 18 October 2010 08:17, Catherine Davey <[log in to unmask]> wrote:
> Hi Eugene! How are you? I'm guessing your post-doc (if that's what it is still?) must be going very well if you're still with the fsl team? Thanks for the response. And thanks Mark also.
Going great guns ;-)!
> I guess the question is most relevant to low pass filtering (since these frequencies have the longest period), but it's relevant to all types. The Matlab filter function (and spm_filter), require time series lengths greater than 3 times the number of coefficients to cater for initialisation at the edges. I was surprised at the requirement as I had expected what Mark wrote - i.e. just restrict the filter for edge voxels. The fact that the matlab filter function doesn't do this made me wonder if there are significant adverse effects in not doing so and, if so, how FSL overcomes these. (I've noticed that for very low cut-offs people occasionally set the relevant frequencies to 0 in the FFT of the time series, but these sharp cut-offs can obviously have adverse side effects, so this isn't a great alternative.)
Depends on the application - I think the basic techniques we've
described are adequate for our applications.
> Also, I wondered if FSL restricted the minimum cut-off frequency to ensure, for example, at least 1 cycle of each frequency, or something of that nature? The paper you referred me to (thanks - it was very useful) suggests a minimum of 2 cycle lengths of a periodic experiment, for example. Perhaps the same applies to other frequencies of interest?
I'm not exactly sure of your application.. for FEAT we the cuttoff
frequency be > 1 cycle length.
Hope you're doing well!
Eugene
> Thanks again. You've been very helpful!
>
>
|