Hi,

The median is more robust to outliers and artefacts, so tends to be a better choice compared to the mean.  In clean datasets they are very similar, but when there are outliers and/or artefacts then the mean can be less representative of the typical tissue compared to the median.

All the best,

Mark

On 19 Apr 2018, at 02:28, huyang <[log in to unmask]> wrote:

Hi Mark,

Thank you for your kind help! In Melodic or Feat GUIs, it seems that grand median scaling is actually used. In other words, the 4D median is scaled to 10000, instead of the 4D mean. I am just wondering is there any special purposes or advantages?

Best,
Yang Hu

At 2018-04-18 22:30:51, "Matthew Webster" <[log in to unmask]> wrote:

Hi Yang,

The ing option considers voxels > 0, you might need to add "-l 0Ħħ to the fslstats call to get matching results.

Kind Regards

Matthew 

--------------------------------

Dr Matthew Webster

FMRIB Centre 

John Radcliffe Hospital

University of Oxford

On 18 Apr 2018, at 02:32, huyang <[log in to unmask]> wrote:

Hi Mark,

My input func image is not masked so that there are no voxels with zero time series. -M and -m options would be equal in this case. I tried both options and the inconsistency mentioned before still exists.

Best,
Yang Hu

At 2018-04-18 05:28:52, "Mark Jenkinson" <[log in to unmask]> wrote:

Hi,

It is likely to be the masking that is the difference.

Try fslstats -M instead of fslstats -m

All the best,

Mark

On 17 Apr 2018, at 07:48, huyang <[log in to unmask]> wrote:

Dear FSL experts,

As for grand mean scaling, I find that there is a difference between using fslmaths -ing option and manual computation. For example: (1) fslmaths func -ing 10000 func_norm
               (2) Mean=$(fslstats func -m); fslmaths func -mul 10000 -div $Mean func_norm
The above two methods would get unequal results and the latter one could scale grand mean to 10000. I wonder how the -ing option is implemented and why?

Best,
Yang Hu