Hi Aize,
I think the hippocampal gray matter is so dark in your images that it is very difficult to distinguish from CSF. Do you have a lot of data collected with these parameters? If not, we probably could suggest some better ones.
cheers, Bruce
On Tue, 18 Aug 2009, caoaize wrote:
Hello Bruce,
The sequence we used is called Turbo field Echo, I need to check it again.
I checked our scanning protocol, found that we did have marker on the subjects' forehead. I generate the right hippocampus from freesurfer, overlay with orig.nii (from orig.mgz), the overlay shows that the subject's right side is the right side in freesurfer's statistical report. It means that our left side is bigger than the right side. This result consitently appears on our other subjects data.
I extract the hippocampus left side mask in this way:
mri_binarize --i aseg.mgz --match 17 --o hippo_left.nii
The label for right side is 53. Anything wrong for what I used?
Also, I used asegstats2table --subject s1 s2... --meas volume --tablefile stats.txt to get the statistical report.
If we accept the low corelation between manual and auto segmentation, how can we explain that the left side is bigger than the right side?
I hope your help. Thanks a lot.
Best Regards,
Aize
Date: Mon, 17 Aug 2009 13:51:26 -0400 From: fischl@nmr.mgh.harvard.edu To: caoaize@hotmail.com Subject: RE: [Freesurfer] (no subject)
wow, I don't think I've ever seen a T1-weighted MR with gray matter this dark. The gray/csf contrast is very low. What kind of sequence is it?
On Tue, 18 Aug 2009, caoaize wrote:
Hi, Bruce,
Here is the aseg.mgz and norm.mgz files. Thanks.
Aize
Date: Mon, 17 Aug 2009 13:27:07 -0400 From: fischl@nmr.mgh.harvard.edu To: caoaize@hotmail.com Subject: RE: [Freesurfer] (no subject)
Hi Aize,
I need the aseg.mgz as well (and probably the norm.mgz). Differences in the definitions of what is hippocampus and what isn't will reduce the correlation coefficient. They should still be somewhat correlated, but even a perfect pair of segmentations under each protocol for each subject will not have a perfect correlation, so it's hard to assess what it means, since the protocols are different. Bruce
On Tue, 18 Aug 2009, caoaize wrote:
Our data is 3T, the manual segmentation follow the McConnell protocol described in "Volumetry of hippocampus and amygdala with high-resolution MRI and three-dimensional analysis software: minimizing the discrpancies between laboratories". I do not know what is the difference between yours and theirs. But, if both method are consistant, then the correlation coeffienct should be high. Here is one image that has bad segmention on happocampus.
I am appreciate for your help.
Aize
Date: Mon, 17 Aug 2009 11:52:19 -0400 From: fischl@nmr.mgh.harvard.edu To: caoaize@hotmail.com CC: nicks@nmr.mgh.harvard.edu; freesurfer@nmr.mgh.harvard.edu Subject: RE: [Freesurfer] (no subject)
it lost interior voxels? That doesn't sound right. Can you send us an image? If your manual protocol is different than the one we used then the DICE overlap is somewhat irrelevant.
cheers, Bruce
On Mon, 17 Aug 2009, caoaize wrote:
> > Hi, Bruce, > > > > I am not clear what manual segmentation protocol we are using, it was done by a trained RA. Currently we just calculate the volumes correlation between manual segmentation and the auto one, but not your DICE method. Our aquisitions are: Philips 3D sense, T1TFE, TR 7.9ms. > > > > In 27 subjects, we have two hippocampus volumes from Freesurfer auto segmentation around 1700, it keeps the edge, but lost lots of inside voxels. I am thinking to manual edit these two volumes, but do not know what happened. > > > > Aize > >> Date: Sat, 15 Aug 2009 09:14:30 -0400 >> From: fischl@nmr.mgh.harvard.edu >> To: caoaize@hotmail.com >> CC: nicks@nmr.mgh.harvard.edu; freesurfer@nmr.mgh.harvard.edu >> Subject: RE: [Freesurfer] (no subject) >> >> Hi Aize, >> >> you are using the CMA manual segmentation protocol? We have not found a >> bias in volumes, and our DICE is usually close to .9 for hippocampus. >> What acquisitions are you using? >> >> cheers, >> Bruce >> On Sat, 15 Aug 2009, caoaize wrote: >> >>> >>> Hi, Bruce, >>> >>> >>> >>> Our segmentation protocol is the same one as Freesurfer's auto segmentation, I used recon-all -s subject -all, command. I do not know in this protocol, whether fimbria and alveus are included. We do not have skin fiducial to tell right or left side. The interesting thing the mean volume for auto is less than manual segmentation. >>> >>> >>> >>> Aize >>> >>> >>> >>> >>>> Subject: Re: [Freesurfer] (no subject) >>>> From: nicks@nmr.mgh.harvard.edu >>>> To: fischl@nmr.mgh.harvard.edu >>>> CC: caoaize@hotmail.com; freesurfer@nmr.mgh.harvard.edu >>>> Date: Fri, 14 Aug 2009 18:53:41 -0400 >>>> >>>> just to add a factoid, from the ADNI data set, 793 subjects processed by >>>> freesurfer: >>>> Volume Mean: StdDev: >>>> ------------------- ----- ------- >>>> Left-Hippocampus 3184.335 604.930 >>>> Right-Hippocampus 3233.753 621.896 >>>> >>>> seems pretty symmetric, with the right being slightly larger (but within >>>> std). this subject set is a mix of controls, MCI and AD patients. >>>> >>>> n. >>>> >>>> On Fri, 2009-08-14 at 17:53 -0400, Bruce Fischl wrote: >>>>> is your segmentation protocol the same as the one we base our automated >>>>> segmentation on? For example, are fimbria and alveus considered part of >>>>> the hippocampus? >>>>> >>>>> As far as left/right switches go, do you have any fiducial markers (e.g. >>>>> vitamin E tablets) in the images? >>>>> >>>>> cheers >>>>> Bruce >>>>> On Sat, 15 Aug 2009, caoaize wrote: >>>>> >>>>>> >>>>>> Hi, there, >>>>>> >>>>>> >>>>>> I have some problem for the hippocampus segmentation results. >>>>>> >>>>>> 1. My structural data are originally in par-rec file. I used r2agui to >>>>>> convert them into .nii format, then use recon-all -i to convert the .nii >>>>>> data into freesurfer .mgz format. Based on these operations, I run >>>>>> recon-all -all on all my data, the statistical report show that the mean >>>>>> left hippocampus volume is bigger than the right one, while literature >>>>>> shows that the right one should be bigger. This problem continuously >>>>>> appeared in our different studies. I am not sure whether it is r2agui's >>>>>> problem, which changed the oritation of the structural data. Do you have >>>>>> any idea? >>>>>> >>>>>> 2. After autosegmentation, we did a statistical comparison analysis >>>>>> between automatic segmentation and manual segmentation, and the >>>>>> correlation is quite low, less than 80%, which is not acceptable for us. >>>>>> >>>>>> 3. Can I edit the auto segmentation results in Freesurfer if we think the >>>>>> auto segmentation is not good enough in some area? >>>>>> >>>>>> I will be really appreciated for your help. >>>>>> >>>>>> Aize >>>>>> >>>>>> >>>>>> _________________________________________________________________ >>>>>> Messenger°²È«±£»¤ÖÐÐÄ£¬Ãâ·ÑÐÞ¸´ÏµÍ³Â©¶´£¬±£»¤Messenger°²È«£¡ >>>>>> http://im.live.cn/safe/ >>>>> _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer >>>> >>> >>> _________________________________________________________________ >>> ÿÿWindows Live ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿMessengerÿÿ >>> http://www.windowslive.cn > > _________________________________________________________________ > ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ > http://www.microsoft.com/china/windows/windowslive/products/photos-share.asp...
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ http://www.microsoft.com/china/windows/windowslive/products/photos-share.asp...
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ http://www.microsoft.com/china/windows/windowslive/products/photos-share.asp...
MessengerÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿMessengerÿÿÿÿÿÿ http://im.live.cn/safe/