Dear Rosalia,
I have not used DARTEL, but I can offer some general pointers on VBM-style analysis using images with stroke lesions. In our lab we use the actual T1-weighted image for correlations with cognitive scores, rather than the segmented image. Depending on your question, it is the correlations with cognitive scores that will be most informative. Comparisons between patients and controls should give you a result equivalent to a lesion frequency map, just showing which regions are most likely to be damaged in your sample of patients.
There are a number of papers about our method, but you will find them cited here:
Tyler et al. (2011) Brain http://dx.doi.org/10.1093/brain/awq369
We argue that, although the T1-weighted signal varies with tissue type, the variability between subjects will reflect the lesion, provided normalisation is good and you control for overall image intensity. Segmentation gives a cleaner, more abstracted measure of tissue integrity, but it may not detect subtle changes present in the original image. You must be sure you understand the transformations involved to know how effects may be missed or introduced artificially. In particular, segmentation is not designed to work with images with large lesions, so you need to be extra sure you have dealt with the lesion.
We use strict warping regularisation rather than cost function masking when normalising lesioned images (using unified normalisation), since this gave better performance in Crinion et al. (2008) Neuroimage http://dx.doi.org/10.1016/j.neuroimage.2007.04.065
If you want to use the segmented image, Seghier et al. (2008) published a method for segmenting lesioned images using a fourth tissue class to capture damaged voxels. This was intended for lesion identification, but could be adapted to generate grey matter images for VBM.
Seghier et al. (2008) Neuroimage http://dx.doi.org/10.1016/j.neuroimage.2008.03.028
Hope this helps,
Paul
On Fri, 27 Jul 2012 13:23:11 +0200, Rosalia Dacosta <[log in to unmask]> wrote:
>Dear SPM experts,
>
>I think it would be better if I explain to you what I have done with this
>image:
>
>1. New segment (asking for deformation fields)
>
>2. Zeroed lesion mask and take c1 (from new segment) and fslmaths with the
>zeroed mask.
>
>3. Dartel tools --> create warped image applying the deformation fields to
>my c1_with_lesion_mask image
>
>4. Modulation --> I am trying to modulate with dartel my warped_c1 image
>(instead of rc1)
>
>5. Normalization and smoothing.
>
>*I am having problems with numbers 4 and 5.
>*
>I hope this helps to understand better what I am trying to do. I know it
>can be very difficult to understand by mail.
>
>Best regards,
>
>Rosalia.
>
|