"this is now where our development effort is focused these days" should have read "this is NOT where our development effort is focused these days"
On Oct 25, 2011, at 9:33 AM, Donna Dierker wrote: > Sorry I don't have time to read this more carefully, but I'm swamped this > week. > > But the quick scan leads me to believe this is an issue with you trying to > segment monkey data at a resolution above 0.5mm. > > The problem is that some of the routines (e.g., especially > eye/skull/hindbrain removal) depend on the number of slices away from the AC > something is. If the number of slices is twice what it expects, it won't > work. You can turn off eye/skull removal, but as you already know, > de-checking hindbrain just makes it fail. > > Caret's segmentation has its limits, and this is now where our development > effort is focused these days. Sorry. > > I know some people have gotten Freesurfer to segment monkey data, but I > suspect there are tricks/tweaks, and I do not know them. I don't know how > the talairach.xfm stuff (and that which depends on it) works, for example. > Obviously MNI305 won't work as a target. > > > On Oct 25, 2011, at 7:54 AM, wolf zinke wrote: > >> Hi, >> >> I had a similiar question a while ago, that was related to your issues. At >> this time there was no 64 bit version of caret available, and hence I ran >> into memory troubles with a high resolution monkey file (do you use the 64 >> bit version of caret?). >> >> However, Donna Dierker gave me some pointers, why it would be problematic to >> use a voxel size different from 0.5 mm. I am not sure, if this is still >> true, but that might be a reason for your problem. I myself had also the >> impression, that for the segmentation a resolution of 0.25 mm would be very >> beneficial. >> >> I hope that this information helps, >> wolf >> >>> Hi, >>> >>> Thanks for the clarification. Currently,I am running caret on the 0.5 mm >>> resolution and it overall gives good results, but fails in occipital >>> regions. However, with your information I have a good reason to stick to >>> the 0.5 mm resolution - makes the manual correction faster anyway. >>> >>> thanks for the reply, >>> wolf >>> >>> >>> >>> On 01/28/2010 04:03 PM, Donna Dierker wrote: >>> Setting aside the memory/64-bit question, there are assumptions built into >>> the SureFit algorithm that assume voxdims for monkeys around 0.5-0.75mm >>> cubed. For example, there are routines in the hindbrain removal that are >>> based on number of *slices* from the AC, and if you double the resolution, >>> those will be off by a factor of two. In short, it will fail. >>> >>> Try downsampling to 0.5 and making sure you crop to left and right hems. >>> If the problems persist, upload your anatomical volume here: >>> >>> http://pulvinar.wustl.edu/cgi-bin/upload.cgi >>> >>> On 01/28/2010 07:30 AM, wolf zinke wrote: >>>> Hi, >>>> >>>> Sorry that I did not reply directly to this thread, but I did not find any >>>> option for this reply. >>>> >>>> Is there a reason why caret is not build for 64 bit systems? I tried to >>>> run a segmentation on macaque data with 0.25 mm voxel size, hoping to get >>>> better results due to the resoltion. However, Caret threw an error about >>>> insufficient memory, which first puzzled me since the PC got 32GB. But >>>> than I realized that due to the 32 bit, Caret is not able to address more >>>> than 4GB of the RAM, right? >>>> >>>> cheers, >>>> wolf >> >> >> On 25/10/11 08:30, Colin Reveley wrote: >>> I wonder if the bits of spine and the affine to to (deskulled, upsampled) >>> F99 are an issue. >>> >>> I don't recall, but it's hugely likely I tried with rigid body too, i.e. >>> without the bit cut off at bottom. >>> >>> On 25 October 2011 00:53, Colin Reveley <[email protected]> wrote: >>> >>> Hello. >>> >>> I have macaque data that is 0.25mm. I like that. I can do things with it >>> that are more than cosmetic. >>> >>> The data was taken with a fancy brukker, and the contrast is very good from >>> the sequence used. so good I wonder if it's a problem (it's a FLASH_MTR - >>> it does correlate to T1 really closely, but contrast GM-WM is clearer. and >>> there may be differences.) >>> >>> Thus far, I've been downsampling to 0.5 to make CARET surfaces. >>> >>> I'm beginning to suspect that, for my project, there is profit in a surface >>> made at 0.25, with many nodes. >>> >>> what I am interested in is the really quite small region (in absolute >>> terms) that was the subject of the paper by lewis and van essen in 2000. >>> >>> even though the F99 atlas does not have 300,000 nodes the paint, border and >>> metric data are scalable and my own data would indeed support a hi-res >>> surface, and benefit from it. >>> >>> I've got RAM. >>> >>> but I never managed to get far with 0.25. >>> >>> segmentation fails with hindbrain at any resolution below 0.5. >>> >>> I didn't mind. But now I think (I really do) I have a good reason to seek >>> surface construction directly from my structural data at 0.25mm. >>> >>> So: is it possible? caret_command ... -res=X ? >>> >>> my data is ex-vivo. And probably no more than 1% of nonGM or nonWM voxels >>> are nonzero. no ventricles. nothing. I did that. A mistake maybe. >>> >>> If I segment at 0.5, upsample to 0.25 and generate a surface with my data >>> it works. CARET can make the surface. >>> >>> but segmentation does not work. >>> >>> appreciate help. >>> >>> Colin >>> >>> >>> _______________________________________________ >>> 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 > > > _______________________________________________ > 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
