On 10/15/2009 03:28 PM, Taosheng Liu wrote: > Hi Donna, > Thank you very much for your quick reply. My current application is > on a small sample of subjects (n<8), and we just want to average some > individual maps in the PALS atlas space. This is a supplementary > analysis to the main ROI-based analysis, which is something I do most > of the time, so my question may sound naive. Please bear with me :) > Thanks for the pointer about the menu. I can now inflate my own > surface (I didn't search hard enough, I guess). I'm going to follow > the tutorial to see how the whole thing works. In the mean time, I > have a few more questions. > 1. So what's the purpose of smoothing the medial wall? Do I need to > do this given my purpose? Is there some instruction on how to do this? > I don't seem to see this step in the tutorial. This step isn't in the tutorial, because it normally gets done automatically during flattening, which was a precursor step to registration at the time the tutorial was written.
The purpose of smoothing the medial wall is to reduce the effects of variability in segmentation of non-cortical medial wall on downstream processes like registration. David once tested registration with and without medial wall smoothing, and the effect was larger than he had anticipated. Hence, we always smooth the medial wall. (We typically use an "OPEN" topo file for generating surface area measurements, which means the medial wall is excluded from that measure, so that is not the reason.) It isn't hard to do: 1) Draw a border around the medial wall (Layers: Borders: Draw borders). 2) Use Surface: Region of interest menu to select the nodes inside the border. (Hopefully the GUI doesn't still need this to be on a flat surface, but eventually I will be urging you toward scripts anyway.) Assign paint of MEDIAL.WALL. 3) Switch to fiducial surface in main window and do Surface: Geometry: Smoothing: Landmark Constrained Smoothing. 4) Save resulting coord file. Then you would use Surface: Geometry: Generate inflated and ellipsoid from the smoothed medial wall fiducial. But there is one problem with this process. It requires a closed medial wall border, while the landmark registration needs two open segments (dorsal and ventral). The closed one should really be the same as connecting the two open segments using the shortest line between the two at each end. The postborder.sh script does exactly that, and it does the medial wall smoothing above. It does other things you will get tired of doing 16 times (n=8 times two hemispheres per subject). And it does them more reliably than someone doing this manually would. I couldn't find in the GUI how to join two open border segments into a closed one. Maybe it's there somewhere, though. > 2. I read the link you sent me. I'm interested in drawing the > landmarks without using the flat surface. Do you recommend this? I > don't really need to flatten, I guess. It'll be fine to visualize the > final result on a inflated surface. The instruction > (http://brainvis.wustl.edu/help/landmarks_core6/landmarks_core6.html/) > seem to suggest it's less accurate to draw it on the spherical > surface, which I agree; it's hard to know what is what there. But can > I draw it on the very inflated one? Seems it'll be easier to draw it > there. You can't really draw the sylvian border on the very inflated surface; there's still too much invagination too it. Ditto on the medial wall ventral. The ellispoid can be easier, but not on the medial wall ventral. The sphere is the best alternative to the flat, in my view. But in your shoes, I'd invest time in getting the scripts to work; let auto-landmarks get you 90% there; and then just have to tweak the borders using Layers: Borders: Border Update (which is downright fun to use). > 3. One of the links mentioned that the volume should be normalized > to standard space. I generally don't do any normalization. Do you > think it's a problem for my purpose? I'm thinking as long as I > register each subject's brain surface to the atlas surface, that > should be ok. It'll be a bit hard for me to do this as my software > doesn't currently support normalization (again, coming from a > ROI-based tradition). Auto-landmarks won't work if the surface isn't in 711-2B stereotaxic space. (And getting it there from MNI isn't hard.) Drawing landmarks will be un-fun and quite dicey if the surfaces are not at least in AC-PC alignment. > > In general, I would like to have relatively high accuracy in my > spherical registration. And given my small sample, I don't mind doing > things manually. Doing things manually often decreases accuracy, but not always. > Could you advise the best thing to do? Sorry some of the questions > might be obvious if I follow through the tutorial but I'm just a bit > overwhelmed by the software right now (it's an awesome software, > that's why I'm trying to learn it). This sounds like a bit of a side project for you, and only you can determine whether it is worth doing. My honest opinion is that doing this manually would prove more work than investing in getting the scripts to work. Yes, even though you have less than eight subjects. If you had three or less and blew off medial wall smoothing, my answer might change. But even six is a lot of surfaces to process manually. But getting the scripts to work does mean spatially normalizing your surfaces, which typically means using something like FSL's flirt or Avi Snyder's imgreg to compute an affine transform to register your anatomical volume to some template (e.g., avg152T1 or Wash U's 711-2*). > Thank you so much, > > --taosheng > > > [email protected] wrote: >> Send caret-users mailing list submissions to >> [email protected] >> >> To subscribe or unsubscribe via the World Wide Web, visit >> http://brainvis.wustl.edu/mailman/listinfo/caret-users >> or, via email, send a message with subject or body 'help' to >> [email protected] >> >> You can reach the person managing the list at >> [email protected] >> >> When replying, please edit your Subject line so it is more specific >> than "Re: Contents of caret-users digest..." >> >> ------------------------------------------------------------------------ >> >> Today's Topics: >> >> 1. Caret newbie question:spherical registration (Taosheng Liu) >> 2. Re: Caret newbie question:spherical registration (Donna Dierker) >> >> >> ------------------------------------------------------------------------ >> >> Subject: >> [caret-users] Caret newbie question:spherical registration >> From: >> Taosheng Liu <[email protected]> >> Date: >> Thu, 15 Oct 2009 10:12:54 -0400 >> To: >> [email protected] >> >> To: >> [email protected] >> >> >> Hello, >> I'm new to Caret and I'd like to run spherical registration to the >> atlas and do group analysis. Most of my analysis is done in custom >> Matlab codes. I use a combination of FreeSurfer and SurfRelax (a less >> used software) to do segmentation and surface recon. >> I'm able to output my surface to Caret fomat, i.e., I can generate >> topo and coords files for a folded hemisphere, and open/visualize >> them in Caret. I tried to follow the Caret_5.5_Tutorial_Segment.pdf >> for doing registration, but I'm currently stuck, because it seems to >> require flattened/inflated/spherical surfaces of my subjects. >> I'm imagining that I can start with my folded surface and do all >> those things, and then follow the tutorial to do spherical >> registration? My question for now is simply: is this the right >> approach? If so, can someone tell me how to start with a folded >> surface to generate the Inflated, Ellipsoid surfaces, etc., that are >> normally generated during the segmentation process in Caret? (e.g., >> the surfaces shown in Fig. 34 in the tutorial) I looked in the menus >> but couldn't find anything. >> Thank you very much, >> >> --Taosheng Liu >> >> >> >> ------------------------------------------------------------------------ >> >> Subject: >> Re: [caret-users] Caret newbie question:spherical registration >> From: >> Donna Dierker <[email protected]> >> Date: >> Thu, 15 Oct 2009 10:53:53 -0500 >> To: >> "Caret, SureFit, and SuMS software users" >> <[email protected]> >> >> To: >> "Caret, SureFit, and SuMS software users" >> <[email protected]> >> >> >> Taosheng, >> >> I don't think there are any tutorials for doing what you want to do. >> We had a tutorial for registering Freesurfer data, but it's pretty >> outdated. We have scripts that are more helpful for that, and you >> may end up adapting those. >> >> It is possible to generate inflated, very inflated, ellipsoid, and >> spherical surfaces using Surface: Geometry: Generate inflated (etc.) >> from fiducial. Then in principle, you could use these inputs for >> registration purposes. >> >> In practice, we smooth the medial wall before registering our >> subjects, which means drawing a border around it (or using >> auto-landmarks to do so); smoothing just the encircled nodes; and >> then regenerating the inflated, very inflated, and sphere using the >> fiducial with the smoothed medial wall. >> >> Depending on your purposes and how many subjects you have, you could >> skip this step. >> >> It used to be that you could draw borders only on 2D surfaces (i.e., >> flat). But now that you can draw on 3D surfaces, flattening can be >> bypassed. >> >> Have a look at the scripts described in this post: >> >> http://www.mail-archive.com/[email protected]/msg01831.html >> >> Even if you don't think you need scripts for your purpose, at least >> you can see the steps we use to register Freesurfer-generated data, >> and if there is something you don't understand, you can ask about it. >> >> Donna >> >> On 10/15/2009 09:12 AM, Taosheng Liu wrote: >>> Hello, >>> I'm new to Caret and I'd like to run spherical registration to the >>> atlas and do group analysis. Most of my analysis is done in custom >>> Matlab codes. I use a combination of FreeSurfer and SurfRelax (a >>> less used software) to do segmentation and surface recon. >>> I'm able to output my surface to Caret fomat, i.e., I can generate >>> topo and coords files for a folded hemisphere, and open/visualize >>> them in Caret. I tried to follow the Caret_5.5_Tutorial_Segment.pdf >>> for doing registration, but I'm currently stuck, because it seems to >>> require flattened/inflated/spherical surfaces of my subjects. >>> I'm imagining that I can start with my folded surface and do all >>> those things, and then follow the tutorial to do spherical >>> registration? My question for now is simply: is this the right >>> approach? If so, can someone tell me how to start with a folded >>> surface to generate the Inflated, Ellipsoid surfaces, etc., that are >>> normally generated during the segmentation process in Caret? (e.g., >>> the surfaces shown in Fig. 34 in the tutorial) I looked in the >>> menus but couldn't find anything. >>> Thank you very much, >>> >>> --Taosheng Liu >>> >>> _______________________________________________ >>> caret-users mailing list >>> [email protected] >>> http://brainvis.wustl.edu/mailman/listinfo/caret-users >>> _______________________________________________ caret-users mailing list [email protected] http://brainvis.wustl.edu/mailman/listinfo/caret-users
