Hi Venkat,<div><br></div><div>I think you can use the mean value to compare across people.</div><div><br></div><div>As you may already know, you can get two types of reports; an overall mean of all pixels in the 3D tract volume and the slice-by-slice report. It the tract of your interest, in your case CST, travels linearly and is more or less perpendicular to one of the planes (for CST, it is perpendicular to axial slices), the slice-by-slice report would be informative. If you use the whole-tract report, not the slice-by-slice, you may also want to try some metrics other than the overall mean, such as median (as KRT suggested) or histogram. You can save the tract as a raw 3D file, load it to RoiEditor as an object, also load an FA map, apply the object as an ROI to the FA map, and then you can save FA values of all individual pixels in the object (ROI), from which you can create a histogram by Excel. This could be an interesting approach for fibers like corpus callosum, uncinate, or SLF, which make a large turn and slice-by-slice report may not be a good idea.</div>
<div><br></div><div>When you use slice-by-slice report, you have to normalize the slice locations across subjects. For example, the CST at pons could be slice #12 in Subject A and #15 in Subject B. The total length of the CST of Subject #A could be 35 slices while that of Subject #b could be 38 slices. This requires "spatial normalization".</div>
<div><br></div><div>What we did in old days is, load all slice-by-slice reports from multiple subjects to an Excel sheet, slide the whole column up and down to make sure that the CST at the pons of all subjects align in the Excel sheet. This is the same as rigid translation to adjust the slice locations. Even if the CST locations are now aligned at the pons level, This doesn't guarantee that the CST is aligned in the motor cortex because the CST lengths are different among subjects. Then you can stretch or shrink the Excel column of each subject so that everybody has the same column length. This is the same as linear alignment using two-point scaling.</div>
<div><br></div><div>While all the location normalization is performed using Excel in this example, you can also do image normalization in the imaging domain. For this, you need to use DiffeoMap (or other kinds of brain normalization software such as SPM and FSL). You can do fiber tracking in the native data. Then linearly normalize all subject data to the MNI space, from which you can get slice-by-slice report in a normalized manner. All slice locations and fiber lengths are the same in Excel, without further modifications.</div>
<div><br></div><div>You can also do normalization of images first to the MNI space and then do fiber tracking in the MNI space. If you want to do this, you have to normalize your tensor data.</div><div><br></div><div>If we push this logic further, then we can ask, "do we need tractography?". </div>
<div><br></div><div>> Suppose one location of CST, say, at the cerebral peduncle (CP) level, is affected and has low FA values in a patient population.</div><div><br></div><div>> If we do CST tracking and get an overall averaged FA, we can say, "it's not a sensitive way because we mix all pixels from other normal locations". This is a valid logic.</div>
<div><br></div><div>>Then we say, "ok, let's look at slice-by-slice report. This should show low FA at the CP level." The only problem is, the CP level slices are not consistent across subjects. So, for each subject, we have to identify slice #. To make sure that the low FA happens only at CP, we want to compare FAs at pons level, internal capsule level, corona radiata level, etc. Then we have to identify slice # of these multiple locations for all subjects. That's a lot of work.</div>
<div><br></div><div>> So, what we usually do is to use one of the image "normalization" tools that can adjust the location, rotation, and size of the brain across subjects. Then all slice levels are the same. We can do slice-by-slice comparison across subjects.</div>
<div><br></div><div>> Well, however, if all brains are well normalized, we can do voxel-by-voxel comparison across subjects without fiber tracking. If patient populations have consistently low FA at the CP, the voxel-by-voxel comparison should show it anyway. This is the logic for voxel-based analysis (VBA), which is a widely used technique. </div>
<div><br></div><div>> There could be two reasons why you want to use tractography over VBA. First, VBA always have normalization error. In this case, if the CP structures are not well aligned after normalization, we may fail to detect the FA loss. Using tractography in each subject, we know that we are accurately comparing CST vs CST (or CP vs CP). In this sense, we can call tractography a "registration tool", with which we can define voxels that belong to the CST in all subjects accurately. Another point is that tractography groups pixels. If the grouped pixels share the same property, such as reduced FA, it can provide much higher SNR compared to voxel-by-voxel comparison. In this sense, you can call tractography a "spatial filter". SNR is always a problem in VBA and we usually group pixels using isotropic filters. Tractography is a structure-specific spatial filter.</div>
<div><br></div><div>> One counter argument is, tractography is not a stable operation. It requires subjective ROI delineation. Also, it is sensitive to noise; if you scan the same patient twice, the tractography results are not the same. So there is a downside of tractography. This is a precision (reliability) issue.</div>
<div><br></div><div>> Another approach we often use is Atlas-based analysis (ABA), which is also based on the whole brain normalization, followed by whole-brain segmentation by applying a pre-defined segmentation file. This pre-defined segmentation file is another kind of structure-specific spatial filter, defining which pixels should be grouped.</div>
<div><br></div><div>> These methods are all related and have pros & cons.</div><div><br></div><div>Hope this gives you a big picture about DTI data analysis.</div><div><br></div><div>Susumu</div><div><br></div><div>
<br><br><div class="gmail_quote">On Wed, Aug 15, 2012 at 11:05 AM, Rajagopalan, Venkateswaran <span dir="ltr"><<a href="mailto:rajagov2@ccf.org" target="_blank">rajagov2@ccf.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div>
<div><font color="#000000" face="Arial">Thanks, Is it ok to take the mean of this mean value (if the data meets assumptions of t-test and you want to do t-test) across subjects when you perform stats</font></div>
<div><font face="Arial"></font> </div>
<div><font face="Arial">Thanks</font></div>
<div><font face="Arial">Venkat</font></div>
<div><font face="Arial"></font> </div>
<div><font face="Arial"></font> </div>
<div><font face="Arial"></font> </div>
<div><font color="#000000" face="Arial">I would use mean values.<br><br>Atul.<br>On Aug 14, 2012, at 4:53 PM, "Rajagopalan, Venkateswaran" <<a href="http://lists.mristudio.org/mailman/listinfo/mristudio-users" target="_blank">rajagov2 at ccf.org</a><mailto:<a href="http://lists.mristudio.org/mailman/listinfo/mristudio-users" target="_blank">rajagov2 at ccf.org</a>>><br>
wrote:<br><br>Dear All,<br><br>I am calculating changes in FA and other DTI metrics along CST. After tract reconstruction I use statistics option to see FA/other DTI metrics values. The statistics gives me Max, Min, Mean, SD values in each and every slice in my image volume. I am wondering whether I can use Max, value (or only mean has to be used) to study changes between my control and patients.<br>
<br>Thanks<br>Venkat<br><br></font></div>
<p>===================================
<br></p><p class="MsoNormal"><font size="5" color="green"><span style="font-size:18.0pt;color:green"></span></font><font color="navy" face="Calibri"><span style="font-size:10.0pt;font-family:Calibri;color:navy"> </span></font><font size="1" color="green" face="Arial"><span style="font-size:7.0pt;font-family:Arial;color:green">Please consider the environment before printing this e-mail</span></font><u></u><u></u></p>
Cleveland Clinic is ranked one of the top hospitals in America by U.S.News & World Report (2010).
Visit us online at <a href="http://www.clevelandclinic.org" target="_blank">http://www.clevelandclinic.org</a> for a complete listing of our services, staff and locations.
Confidentiality Note: This message is intended for use only by the individual or entity to which it is addressed and may contain information that is privileged, confidential, and exempt from disclosure under applicable law. If the reader of this message is not the intended recipient or the employee or agent responsible for delivering the message to the intended recipient, you are hereby notified that any dissemination, distribution or copying of this communication is strictly prohibited. If you have received this communication in error, please contact the sender immediately and destroy the material in its entirety, whether electronic or hard copy.
Thank you.<br><p></p></div><br>_______________________________________________<br>
mristudio-users mailing list<br>
<a href="mailto:mristudio-users@mristudio.org">mristudio-users@mristudio.org</a><br>
<a href="http://lists.mristudio.org/mailman/listinfo/" target="_blank">http://lists.mristudio.org/mailman/listinfo/</a><br>
Unsubscribe, send a blank email to: <a href="mailto:mristudio-users-unsubscribe@mristudio.org">mristudio-users-unsubscribe@mristudio.org</a><br>
<br></blockquote></div><br></div>