Dear María,

The recommended rule of thumb is three voxels FWHM smoothness, to provide sufficient smooth and approximate the behavior of a continuous random field. However, in many cases the smoothness is higher (Nichols and Hayasaka 2003). Optimal FWHM kernel is difficult to assess since depend also in your scanner and reconstruction algorithm, and I am not aware of many studies exploring the optimal value.

I will recommend you to consider several things before deciding what to do:

- Check the initial voxel size of your images, before being resliced to template voxel dimensions (are you reducing or increasing the voxel size)

- Double check the quality of your template. I consider that a voxel size of 1.6 in z axis is relatively big. As I mentioned earlier, newest MRI templates have a Z dimension of around 0.8 mm (e.g. Nie et al. 2013 or Schwarz et al. 2006).

- Compare with similar studies in humans. In many cases, for voxel sizes of 2x2x2 mm the FWHM is between 8x8x8 and 12x12x12. So, use this values also as a reference.

- Try the analysis with a couple of FWHM values, and check if the results in the clusters are two big or small, etc…

I will suggest you to start testing with (x3 voxel size): 6 x 6 x 24 mm. But again, this depends on the resolution of your scanner and your reconstruction.

Best regards,

David

________________________________
David Vállez Garcia, PhD

[log in to unmask]
University Medical Center Groningen
Nuclear Medicine & Molecular Imaging
Huispostcode: EB50
Postbus 30001
9700 RB Groningen

El 27 nov 2015, a las 17:24, María Pérez <[log in to unmask]> escribió:

Dear David,

Thank you very much for your reply. Yes, previously the fMRI images were scaled by 10. I resliced the images to the template's resolution (1.94 x 1.94 x 8 mm3, the template was scaled by10 already) to calculate later group-ICA. Does the FWHM have to depend on the voxel resolution? Could it be 4 x 4 x 0 mm3 or  4 x 4 x 16 mm3 in the case I mention?

All the best,

María

2015-11-27 16:39 GMT+01:00 David Vállez García <[log in to unmask]>:

Dear María,

Before replying to your question. Why did you reslice the images to 1.94 x 1.94 x 8 mm3? I am also assuming that you are scaling your data x10, otherwise the voxel size for a rat brain makes no sense. Since you have performed a spatial transformation of your original image, you can perfectly reslice your data to a voxel size more “comfortable” (the same than your reference template). In addition, most of the MRI templates have 0.1 or 02mm (real) isotropic voxel size, so your voxel size may not be ideal.

About your question, at least for PET scans of rat brain (which is with what I have more experience), a FWHM of 0.8 - 1.2 mm is quite common. If you are working with MRI with better resolution than a PET, you may prefer to reduce the FWHM.

And finally, it is perfectly possible to perform the analysis of rat/mouse brain data in SPM8 and SPM12 without scaling your data.

Best regards,

David Vállez García


________________________________
David Vállez Garcia, PhD
[log in to unmask]
University Medical Center Groningen
Nuclear Medicine & Molecular Imaging
Huispostcode: EB50
Postbus 30001
9700 RB Groningen

> El 27 nov 2015, a las 15:59, María Pérez <[log in to unmask]> escribió:
>
> Hi all,
>
> After normalising rat images to a rat template and reslice them to 1.94 x 1.94 x 8 mm3 I am wondering how much smoothing would be suitable.
>
> I have been advised 4 x 4 x 0 mm3 or  4 x 4 x 16 mm3 (approx two times the resolution in x, y, z). The doubt is the FWHM in z axis. Please, could you give me some advice and also explain me the advantages and disadvantages of both quantities of smoothing? How does SPM manage the fact that the amount of smoothing is not an exact multiple of resolution?
>
> Thank you very much,
>
> María