 |
 |
This is becoming rather a common question, and it's all to do with how the normalise to MNI space option handles missing data. The fix is to use a mask image, to define those parts of the data that should be included in a second level analysis. The smoothing in the normalise to MNI space option is based on computing a weighted average of the voxels with data in them. After a first level analysis, the contrast images are masked such that voxels outside the brain are set to NaN (not a number), and these NaNs indicate that there is no data available in certain voxels. Therefore, when the contrast images are normalised and smoothed, the weighted average is derived from the voxels under the Gaussian convolution kernel that are not NaNs. Away from the brain, voxels are NaN, so the average is from the closest parts of the brain with signal. This causes the flowery pattern you see. Your first level analyses should produce a mask.img file. These can be normalised to MNI space (along with the contrat images), and a new explicit mask can be generated, eg by using ImCalc and evaluating an expression such as: (i1>0.5) & (i2>0.5) & (i3>0.5) & (i4>0.5) & ... Once you have an explicit mask, then I think the second level analysis should be able to use it. Best regards, -John ps. The part where you overlay results on to individuals would be tricky with data normalised to MNI space. The DARTEL warps are only between the individual subjects and the common average (rather than MNI space). > I used DARTEL to warp all of my functional data to a group-specific template > and did a second level analysis with really nice results. However, now I > want to redo the analysis in MNI space so I can properly identify MNI > coordinates of regions of interest. > > In the analysis I did initially in group-specific space, I did: > 1. Segment anatomicals > 2. Initial Import > 3. Run DARTEL (create Templates) - using grey and white segments > 4. Create Warped (using contrasts from 1st level analysis (con_0001.img) and > corresponding > flow fields u_rc*) > 5. Smoothed the resulting warped contrasts (in group-template space) with > 5mm kernel > 6. Ran 2nd level analysis on smoothed, warped contrasts. > > This looked great, but since I'd like to overlay the results on an MNI > brain, I instead tried to use "Normalize to MNI Space" using the flow fields > I created previously (u_rc*) and > the contrasts I used previously (con_0001.img) as well. Before I did the > analysis, I > looked at the results of these steps to make sure the images looked > reasonable and I'm attaching the results. > > Upper left is the group-specific template. > Upper right is the contrast warped to the group-specific template with DARTEL > Lower left is the contrast normalised using Normalise (Estimate and Write) > Lower right is the contrast normalised using DARTEL's Normalise to MNI Space > (the one I'm worried about) > > Any idea why there are flowery blobs around the contrast when it is > normalised to MNI with DARTEL? I would really like to use DARTEL so I can > invert the transformation and look at the significant group-level regions > back in each subjects' native space, so any help would be greatly appreciated. -- John Ashburner <[log in to unmask]>