Dear Mark,

thank you for your reply !

Actually, I'm using fsl's SUSAN for smoothing, which involves non-linearities. The regular feat cascade also uses SUSAN for smoothing according the logs.

So, if am using SUSAN for smoothing (rather than performing convolution with a smoothing kernel using fslmaths), the order of the smoothing and the high-pass filter operation does matter, correct ?

But do you also think that it is a mistake / introduces bias, if we do high pass filtering prior to SUSAN (rather than after SUSAN as implemented in feat) ?

Thank you for your time !
kind regards
andi

-----Original Message-----
From: Mark Jenkinson <[log in to unmask]>
To: FSL <[log in to unmask]>
Sent: Thu, Aug 2, 2012 9:13 am
Subject: Re: [FSL] high pass filter before smoothing

Dear Andi,

If you do linear flitering and linear smoothing (as we do) then the order does not matter.

These linear operations are interchangeable.

If you want to do some non-linear filtering, then this changes and the order does matter.

I hope this helps.

All the best,

Mark

On 31 Jul 2012, at 21:15, Andi Heckel wrote:

Dear FSL Experts,

I want to denoise fmri data using motion regressors and white matter / CSF masks. I want to perform denoising in native space (after motion correction and unwarping) and before
smoothing, because i would like to avoid the smearing of grey matter signal into the white matter / liquor masks. I guess it makes sense to remove
scanner drifts using a high pass filter before regressing out motion parameters.

I noticed that in the regular feat cascade,  high pass filtering (fslmaths -bptf) is carried out after smoothing, whereas in my customised cascade the filtering is performed
before smoothing.

My question is:
Can I safely perform high pass filtering (to remove drifts) prior to smoothing? Or is it crucial/better to do it after smoothing and -if so- what is the reason behind ?
Secondly, is it necessary to perform high pass filtering before regressing out white matter /CSF signal (since scanner drifts may also be captured by the WM/CSF regressor of no interest) ?

Thank you very much !
Kind regards,
andi