<HTML><HEAD>
<META content="text/html; charset=utf-8" http-equiv=Content-Type>
<META name=GENERATOR content="MSHTML 8.00.6001.18928"></HEAD>
<BODY style="MARGIN: 4px 4px 1px; FONT: 10pt Tahoma">
<DIV>Dr. Mori,</DIV>
<DIV> </DIV>
<DIV>Thank you very much for your help on my last questions. In response to your answer on my questions from Jul 28 (regarding DTI data before/after intervention) - see below for details - I have a follow up question:</DIV>
<DIV> </DIV>
<DIV>I have only used DTIstudio and ROIEditor, and am not too comfortable with Diffeostudio yet. Since you mentioned to try linear transformation I tried the following using AIR in DTIstudio.</DIV>
<DIV>- I have all my processed data saved in analyze format - so I loaded these in DTIstudio.</DIV>
<DIV>- using rigid body transformation with AIR I get really nice results for co-registration of B0 images before & after treatment, but when I use the FA maps and individual val0,1&2 & color maps this (not fully surprising) don't always correlate well. </DIV>
<DIV>- I saved the AirMatrix.txt file from the B0 transformation </DIV>
<DIV>My questions:</DIV>
<DIV>--- How can I transform the other images (FA, val0, val1, val2, color map) using this Matrix???</DIV>
<DIV>--- Can this be done using DTIstudio or will that have to be done in Diffeomap</DIV>
<DIV>--- Is this an acceptable way to do it as long as the B0 images correspond (which they visually do nicely) and is there an easier way to do this using Diffeomap?</DIV>
<DIV> </DIV>
<DIV>Thanks</DIV>
<DIV> </DIV>
<DIV>JMT</DIV>
<DIV> </DIV>
<DIV> </DIV>
<DIV>On Wed, Jul 28, 2010 at 4:01 PM, Jan Tillema <<A href="http://lists.mristudio.org/mailman/listinfo/mristudio-users">Jan.Tillema at cchmc.org</A>> wrote:<BR><BR>><I> I have a similar question.<BR></I>><I> We have dataset of subjects before/after intervention, we have 15 dir DTI<BR></I>><I> and 3D T1 data (GE) with same parameters both times, but orientation not<BR></I>><I> exactly the same (clinical protocol). The brains are not normal at baseline<BR></I>><I> with enlarged ventricles that may have changed minimally (due to treatment,<BR></I>><I> e.g. smaller size). Normalization is not a possibility d/t abnormalities on<BR></I>><I> brain to begin with, we just want to compare before/after. We are<BR></I>><I> investigating the effect on white matter and would like to get best approach
</I>><I> to compare basic measurement before/after. We have manual (blinded) ROI<BR></I>><I> showing differences before/after, but want to make sure we are in the exact<BR></I>><I> same ROI. To co-register seems the best way, but tricky as the structure of<BR></I>><I> the brain is not exactly the same. Any suggestions?<BR></I>><I><BR></I>><I> 1) would it be best to co-register DTI to T1 at both time points and then<BR></I>><I> to co-register T1 in time point A to B followed by change of DTI with same<BR></I>><I> matrix? If so, I assume this would have to be done with DiffeoMap, what<BR></I>><I> would be the steps to do so?<BR></I>><I><BR></I><BR>Unless there is a reason that you want T1-DTI be registered (e.g. ROI drawn<BR>in T1, FA measured in DTI), I think you can directly register DTI to DTI.<BR>Yes, DiffeoMap should be able to do it.<BR><BR><BR>><I> 2) is co-registration to "dangerous" for bias as there is the possibility<BR></I>><I> that the brain is not exactly the same before and after (ventricle size<BR></I>><I> change)<BR></I>><I><BR></I><BR>First you have to decide if you want only linear transformation adjusting<BR>the overall head positions or do non-linear to remove the tissue deformation<BR>due to the surgery. I suspect you want the former (I think the latter is<BR>difficult).<BR><BR>Linear transformation is mostly driven by brain boundary. So, you can assume<BR>that as long as the tissue deformation is mostly inside the brain and the<BR>brain boundary shape remains the same, the linear transformation should be<BR>able to match the two brains. However, we have seen not-so-perfect linear<BR>registration even between two images from the same person. Reason could be;<BR>the initial alignment was way off, there is an intense sinus signal in one<BR>of them, etc. In the end, it is your subjective judgment if the linear<BR>registration is satisfactory. If there is a good quantitative way to judge<BR>the alignment, we would have used such a metric to drive the linear<BR>registration.<BR><BR>In your case, you may want to try the landmark function of DiffeoMap. You<BR>need at least 3 point landmarks to define the corresponding structures<BR>between the two images. Unfortunately there are not many point landmarks<BR>inside the brain. The exceptions are AC and PC. The third point could be<BR>some easy-to-define structures in the brainstem at the mid-sagittal level<BR>such as the junction of the pons and midbrain. If you can clearly find these<BR>3 points, place these landmarks in both images and use rigid alignment.<BR><BR>Let us know if this approach works.<BR></DIV></BODY></HTML>