Hi,
I'm not quite sure what image you are drawing the manual WMLs on - but the usual practice is to do it on the original (native) image. This mask is then inverted (so that there are ones everywhere except inside the WMLs, where it is zero) and is used as a cost function weighting image for the registration of the native image to the MNI152. We would recommend that the registration is done with FLIRT and then FNIRT (both using the cost function weighting image). You would then invert the non-linear warp and use this to transform your lobe mask from the standard (MNI) space to the native space. At this point you could use it in the same space as the WML mask. Alternatively, you could transform the WML mask into the standard space and use the two masks in standard space.
The crucial difference in what I'm describing above and what you've done is the use of the cost function weighting image. This can really help get better registrations for pathological subjects.
All the best,
Mark
On 22 Nov 2012, at 05:16, Ai Qing <[log in to unmask]> wrote:
> Dear FSL experts,
>
> I met a new question in registeration. The following is my protocol:
>
> 1) original data: 3D FLAIR, voxel size 1*1*1mm^3.
> 2) using bet to get rid of skull
> 3) using FLIRT to register it to MNI152_1mm space.
> 4) segment white matter lesions(WMLs) by hand
> 5) draw a standard mask of brain anatomical lobe in the standard MNI152_1mm space.
> 6) add the WMLs mask to the mask of brain lobe, and differentiate the WMLs in different brain lobe.
>
> When I add WMLs mask to the mask of brain lobe, I found some lesions change their original position. For example, WMLs in the corpus callosum may exceed the extent of corpus callosum.
>
> I know all this caused by FLIRT. So I tried to using FNIRT to 3D FLAIR. But I found the distortion of the FNIRTed 3D FLAIR images was so obvious. So I don not think the FNIRTed images is suitable for segmenting WMLs.
>
> Is there a way to make the WMLs(segmented in FLIRTed images) to fit the standard space more accurately?
>
> Thanks all your hard work to solve many problems for us.
>
>
>
|