Hi,
The setup_masks script does not apply any transformations, so you need to take the structural to template warps and apply them directly to the lesion masks in order to get them into the same space as the image being used as the input to randomise. You need a mask in the same space as the image input to randomise, and this will involve the application of a transformation and then the thresholding and binarising that I was describing. Once this is done you can use the setup_masks script with randomise.
I hope this is clearer now.
All the best
Mark
On 29 May 2013, at 15:03, "Papazoglou, Sebastian" <[log in to unmask]> wrote:
> Dear Mark,
>
> Many thanks for your reply.
>
> Unfortunately, I am not sure whether I understand you correctly.
>
> How should I use the transformed and binarised masks other than as input to setup_masks? In our protocol the orginal 1mm masks enter the processing via fnirt’s --inmask option to exclude the lesions during GM template creation. So, I do not see any reason to use the transformed 2mm masks here. I thought they are required to exclude the lesion voxels in the statistical analysis. Or is this last step redundant in our protocol?
>
> Best,
> Sebastian
>
> Am 29.05.2013 um 09:57 schrieb Mark Jenkinson <[log in to unmask]>:
>
>> Dear Sebastian,
>>
>> This issue is really nothing to do with setup_masks per se and really just about what to do with transformed lesion masks in general. I would say that the safest thing to do in this case is to make sure that you exclude the lesions from your calculations, and so you should transform the masks (getting values between zero and one) and then threshold and binarise them. I would choose a threshold near zero, so that you exclude voxels in the 2mm space that have anything more than a minor overlap with the lesion masks in the 1mm space. A threshold of say 0.1 would be appropriate. The analysis would then continue in such a way that only 2mm voxels that had less than a 10% overlap with the lesion mask would be included in the analysis. This would be true whether you used setup_masks or not.
>>
>> I hope that helps.
>>
>> All the best,
>> Mark
>>
>>
>>
>> On 28 May 2013, at 09:47, Sebastian Papazoglou <[log in to unmask]> wrote:
>>
>>> Dear Experts,
>>>
>>> I have an issue with setup_masks.
>>>
>>> We built a VBM template using customised versions of the scripts fslvbm_2 and 3 that account for lesion masks.
>>> The resolution of the original data is 1mm iso. During registration to standard space the resolution is downsampled to 2mm.
>>> Hence we will also have to coregister our lesion masks correspondingly in order to generate proper regressor images
>>> using setup_masks.
>>>
>>> Would you recommend using the interpolated lesion masks (values between zero and one) as input to setup_masks-
>>> or is it mandatory to use thresholded "binary" masks?
>>>
>>> Many thanks in advance.
>>>
>>> Best regards,
>>> Sebastian
>
> --
> Dr. rer. nat. Sebastian Papazoglou
> Post Doctoral Research Fellow
> NeuroCure Clinical Research Center NCRC
> Charité University Medicine
> Charitéplatz 1
> 10117 Berlin
> Germany
>
> Tel. +49 03 450 539797
> [log in to unmask]
|