Dear Anderson,

Thank you for your reply. I experimented around with our data. When I use SUSAN, I get better results for one T1 (mostly in the temporal lobes). I wasn't able to average two T1's filtered with SUSAN because FreeSurfer gave me an error (they did not have the same dimensions) but it seems that one T1 filtered with SUSAN does pretty much the same job as two T1's averaged (but without SUSAN).

I'll have to figure out how to get the same dimensions when I run SUSAN for two T1's (same subject) so I can average them in FreeSurfer and then compare the results with non-filtered averaged T1's.  

I tried to attach some pictures but my message was rejected due to the email size. So I am sending this again without pictures.

Best,

Mihaela

On Fri, Jan 9, 2015 at 4:09 PM, Anderson M. Winkler <[log in to unmask]> wrote:

Dear Mihaela,

I used SUSAN in the past in a certain pretty noisy dataset and the results were quite good. The parameters to use were defined heuristically in a few test cases, trying to be being parsimonious to just minimise the noise a bit while preserving the tissue borders (in particular, the thalamus', which are naturally a bit blurry in certain areas). The exact same parameters were then used for all subjects (200+) in that sample.

Not sure about others' opinions, but I don't see a problem in using.

Note, however, that it isn't a replacement for good image acquisition, especially multiple acquisitions, which still yield better results. Also, it won't dramatically improve movement artefacts on the surfaces (e.g., those common in the temporal lobes).

All the best,

Anderson

On 9 January 2015 at 20:36, Mihaela Stefan <[log in to unmask]> wrote:

Hello,

We have some noisy data and we found that FSL-SUSAN was able to improve the T1 image quality (acquired on GE T3, BRAVO sequence). We want now to use the corrected images in FreeSurfer. Has anyone tried this before?  Is there any reason not to apply a noise-reduction filter before recon-all?

Thanks!
Mihaela