 |
 |
Hello, I think you can safely ignore the cluster of size 3 that is right next to the cluster of size 50. Even if it were statistically significant with standard measures, none of these measures take into account the potential for systematic misregistration, and so I believe it is valid to simply make the interpretation that this is unlikely to be a biologically meaningful result. As for actually calculating cluster-based statistics, at present it is not implemented, as I said. It is conceptually simple to implement a randomisation-based algorithm that could calculate the cluster size and use this as a statistic, however this is not currently possible with our tools as they do not take surface inputs at present. It would also be possible to do a Gaussian-Random-Field-based implementation, but this would be more difficult and still not possible with the current tools. I would stick to making the simple interpretation at present and commenting on this clearly in any manuscript that you prepare. As I said before, even the statistic methods mentioned above cannot deal with systematic mis-registrations, so it may still be necessary to comment on it in this way. All the best, Mark On 8 Dec 2009, at 02:45, duhanjian wrote: > Dear Mark, > Thank you for your quickly reply. > As regard to the cluster threshold, for example, I get 2 > clusters, the cluster size of them is 3 vertices of inflation and 50 > vertices of atrophy respectively. > We can choose a threshold, for example 5 vertices, then, the cluster > of 3 vertices is masked since the cluster size of the cluster is > smaller than 5. We think the cluster smaller than a threshold is > considered to be caused by misregistration and thus should be masked. > Do you think that a cluster of 1 vertex is statistically > significant and meaningful? > I think clusters smaller than a threshold should be masked. So how > to determine the cluster threshold which should be masked? > Thank you very much! > > 2009-12-08 > yours sincerely, > hanjian du > Department of Neurosurgery, Southwest Hospital, Chongqing 400038, > China > > 重庆市西南医院神经外科 > 杜寒剑 > 发件人： Mark Jenkinson > 发送时间： 2009-12-08 07:39:35 > 收件人： FSL > 抄送： > 主题： Re: [FSL] several questions about the FIRST tool > Hi, > The inward vector association with atrophy depends on which way > around your groups were defined. See the documentation for details > on this. > If you did have a small cluster of oppositely oriented vectors next > to a large patch then this may indeed indicate a misregistration > within the surface. It is sometimes quite difficult to get good > registration within the surface as there are not many anatomical > features to go by. It probably depends on what structure and > where in the structure you are seeing this. Definitely worth being > cautious about interpretation though. If the neighbouring patch > is much larger than the opposite cluster then I think the large > patch is probably safe to interpret as "real" but I would not > do so for the small cluster. > The FDR correction that we have at the moment does not > use cluster size. It is theoretically possible to do cluster > correction (actually, fairly easy) but we have not yet > implemented this form of correction. > All the best, > Mark > On 7 Dec 2009, at 15:32, Zhangyuanchao wrote: > > Hi,there, > > > > I wonder whether I can think that it is atrophy if the end point of > > the arrow lies inside the closed surface and that it is inflation if > > the end point of the arrow lies outside the closed surface. > > > > If yes, how could I cope with the situation that there is a very > > small cluster of vertices of inflation near a large cluster of > > atrophy. It seems to me that it might be caused by miscorrespondence > > between corresponding vertices. > > > > Is there any suitable threshold of cluster size after FDR > correction? > > A cluster smaller than the threshold is considered to be caused by > > noises and should be masked. > > How many vertices are considered significant? > > > > Yuanchao > > > > > > > > > > > > --- 09年12月7日，周一, Mark Jenkinson <[log in to unmask]> > 写 > > 道： > > > > 发件人: Mark Jenkinson <[log in to unmask]> > > 主题: Re: [FSL] several questions about the FIRST tool > > 收件人: [log in to unmask] > > 日期: 2009年12月7日,周一,下午5:42 > > > > Dear Hanjian, > > > > This is actually not my personal mailbox - it is the FSL list. > > I have been away from work for a few days. > > Also, I am not familiar with other packages for displaying vtk files > > and I > > do not know if there is a strict standard as to which orientation > (for > > the left-right handedness) they are normally displayed. We display > > things consistently with FSLView, which involves flipping left-right > > for "neurologically ordered" files (positive determinant of the > qform > > or sform matrix). Our vtk coordinates use an internal coordinate > > convention and do not follow the nifti conventions. Hence I am not > > surprised that you see a left-right flip given these two significant > > conventions (coordinates and display in FSLView). So I would > > not worry about this. > > > > The arrows show the direction of the mean change between > > groups. If they are tangential to the surface then the change > > is along the surface and neither inward nor outward. > > > > I would not use fslswapdim to correct registration *errors*! > > It is appropriate to use fslswapdim on your original images > > to either fix labelling problems (A-P,L-R,S-I) or to put the images > > in a more familiar orientation (like the standard templates). > > However, I would *NOT* run it *after* any analysis has been > > done. If you have a registration error then you should try to > > fix it by cropping the image to remove large amounts of neck, > > checking the brain extraction, changing the search space or > > even selecting a different cost function. > > > > I hope this helps. > > All the best, > > Mark > > > > > > > > > > On 4 Dec 2009, at 06:22, duhanjian wrote: > > > > > Dear Mark, > > > I am very sorry to mail this letter to you mailbox directly. > > Beacause I haved sent those questions to the FSL mail box several > > days ago, but nobody answer those questions. So I mail those > > questions to you. Could you do me a favour? > > > I was confused by following the questions. > > > 1.Using FIRST package, I tried to make group analysis in patients > > vs normal controls. I found there was group difference in left > > amygdala and other structures. I tried to visualize the difference > > using vtk by my self. In the uploaded figure, I showed the left > > amygdala using vtk and fslview respectively. To my surprise, they > > are different. It seems to me that my result is a mirror image of > > the result obtained using fslview. In addition, the result obtained > > using fslview is smoother. Since I loaded the same > > NS_L_Amyg1_Fthresh.vtk file, why are the different in that way? > > > > > > 2.In addition, there are arrows which are tangent to the surface > > but the arrow is still outside the surface or inside the surface.Can > > we say that it is atrophy if the arrow lies inside the surface, no > > matter it is pointing inward or tangent to > > > the surface? > > > > > > 3.In order to correct registration errors, I used fslswapdim > > command. What is the effect of fslswapdim? Should I take any > > additional steps when making group analysis if I performed the > > fslswapdim command? > > > > > > By the way, I have uploaded the pictures to http://www.fmrib.ox.ac.uk/cgi-bin/upload.cgi > > . The six-digit reference number for this upload session is 555243. > > > > > > Thank you very much! Mark > > > > > > > > > > > > 2009-12-04 > > > yours sincerely, > > > hanjian du > > > Department of Neurosurgery, Southwest Hospital, Chongqing 400038, > > China > > > > > > 重庆市西南医院神经外科 > > > 杜寒剑 > > > > 好玩贺卡等你发，邮箱贺卡全新上线！ > .