Dear Yu-Shiang,
Could you describe a bit further your rationale for changing the masking
threshold, from 0.8 to -Inf? Here, it seems that you are using an
explicit mask to restrict the analysis mask from the one obtained with
threshold masking. If so, why not leave the threshold at 0.8 but add the
explicit mask? The final analysis mask will be the intersection of the
two masks so that the misbehaving voxels at the bottom of the image will
be discarded.
Best regards,
Guillaume.
On 24/10/2019 17:46, Yu-Shiang Su wrote:
> Dear SPM experts,
>
> I have an identical subject run first-level estimation with and without
> threshold masking (set default 0.8 or set to -Inf and apply explicit
> mask). The resulting images puzzled me as showing in the attachment.
> Comparing images with threshold masking and without threshold masking,
> the beta_0001.nii and con_0001.nii have similar distribution (the
> locations of dark and light cluster seems similar). However, t-map with
> threshold masking have proper distribution, and t-map with no threshold
> masking is totally flat that the value in every voxels is close to zero.
> I check the ResMS image and found that the values in ResMS with
> threshold masking are small (< 1.0). But the values in ResMS without
> threshold masking are very large (around 500 ~ 1500).
>
> It seems that con_0001.nii is not affected (although the value in the
> same voxel have very small difference around .01 ~ .001). So I assumed
> that the second-level should not be affected too much. I still have some
> questions here: (1) What the cause that when I change threshold masking
> to -Inf I would got very huge residuals here. (2) It is any possibility
> that I apply threshold masking to -Inf and still able to get proper t
> map? (3) Would these high residuals affect analysis in second-level?
>
> Thanks so much for your attention. All comments are helpful to me.
>
> Best,
> Yu-Shiang
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
UCL Queen Square Institute of Neurology
London WC1N 3BG
|