Dear Mark,
Hi, many thanks for getting back to me.
Sorry! I don't know either. If it looks good for you and not for me, perhaps we have some local installation issues?
I have uploaded a new file for you via OxFile with some segmentation results from the brain-extracted images. Please take a look and see if you agree and have any ideas.
The name is: For_MARK_from_COLM.tar.gz
Many thanks!!
Best wishes,
Colm.
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Mark Jenkinson
Sent: 25 October 2013 21:05
To: [log in to unmask]
Subject: Re: [FSL] FIRST - bad segmentation for caudate only
Dear Colm,
I don't know why you are still having trouble.
Can you provide an example for me?
All the best,
Mark
On 24 Oct 2013, at 12:20, "McGinnity, Colm" <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Dear Mark,
Thank you very much for getting back to me and your helpful suggestions.
I agree this looks like a registration problem. We have run using brain-extracted version and -b as suggested; unfortunately in our hands the results still look poor, still significantly expanded. Do you have any idea why this could be? I have tried using the bias-corrected image, but this didn't help.
If you have any other suggestions, I would be very grateful to hear them.
Many thanks!!
Best wishes,
Colm.
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Mark Jenkinson
Sent: 21 October 2013 08:00
To: [log in to unmask]<mailto:[log in to unmask]>
Subject: Re: [FSL] FIRST - bad segmentation for caudate only
Dear Colm,
I've had a look at a sample of the images you sent and I believe the problem is probably the initial registration. Did you check the registration results? When I ran standard run_first_all, it produced a very poor registration to the standard (significantly expanded) and therefore it will have got the initial location of the caudate wrong, which has biased the final result.
So to fix this I took the original image, brain extracted it, and used the brain extracted version with run_first_all and the -b option. This worked much better. So try this and see if you are happy with the result.
Remember that it is important to check the registration results each time, as if the registration did not work well then I would not trust the segmentation results.
All the best,
Mark
On 15 Oct 2013, at 15:22, "McGinnity, Colm" <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Hi Mark,
Thanks for your email, sorry for the delay. I too have uploaded some files: they begin ROLAND*A*.hdr/.img and FIRST_vOLD*.nii.gz
Thank you very much in advance for looking at them.
Best wishes,
Colm.
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Miguel Burgaleta
Sent: 14 October 2013 13:08
To: [log in to unmask]<mailto:[log in to unmask]>
Subject: Re: [FSL] FIRST - bad segmentation for caudate only
Hi Mark,
I uploaded some images with their corresponding segmentations. The bad ones are mostly for the right caudate.
The file name is QC_RFP_Upload.tar.gz.
Thanks,
Miguel
On Fri, Oct 11, 2013 at 11:21 PM, Mark Jenkinson <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Dear Colm and Miguel,
I did not end up being able to see images to look at for this so I couldn't help more.
Can you upload some examples to the following website:
https://oxfile.ox.ac.uk/oxfile/work/extBox?id=68312615463381F4C
I don't think I can help more before seeing some data.
All the best,
Mark
On 10 Oct 2013, at 15:01, Miguel Burgaleta <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Hi Colm,
I'm afraid I don't have any news on this... The issue persists and I found no solution so far. I'm just excluding the failed segmentations from my analyses.
It would be great to know if new solutions are available though, as for one of the datasets I'm currently working on the rate of bad segmented caudates is really high (50-60%). Please keep me in the loop.
Best,
Miguel
On Thu, Oct 10, 2013 at 12:42 PM, Colm McGinnity <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Dear Miguel, Mark, (and FSL Experts),
Hope this finds you well. I was just wondering if you or anyone else had managed to find a solution to this caudate problem (see below)? As I believe we are suffering from the same problem with our data.
Hope to hear from you!
Many thanks and best wishes,
Colm.
Thanks Mark for your reply.
Registrations look fine. The non-boundary corrected segmentations don't cover all the caudate though.
I have uploaded a few problematic T1s, ref 277951 (not sure if you want me to upload the whole data set and/or the outputs, please let me know)
Thanks so much!
Best,
Miguel
On Wed, Jan 23, 2013 at 10:50 AM, Mark Jenkinson <[log in to unmask]<mailto:[log in to unmask]>> wrote:
Hi,
I'm not sure what to suggest without seeing the images.
Is the registration sufficiently accurate?
If not then try working on that.
Does the non-boundary corrected image cover all of the caudate?
If so then some other option for boundary correction might work.
If you still can't get a good solution then send us the data (see the Support page in the website for details) and we'll have a look.
All the best,
Mark
On 22 Jan 2013, at 17:04, Miguel Burgaleta <[log in to unmask]<mailto:[log in to unmask]>> wrote:
>
> Hi FSLers,
>
> I think this one is for Mark.
>
> I usually get stunning subcortical segmentations with FIRST, with one exception: the right (and sometimes, left) caudate nucleus. This happens consistently to me across data sets. When I overlay the fast_firstseg.nii.gz over the subject's T1, I often see that an anterosuperior segment of the structure is excluded. I can see this in aprox 15% of the subjects.
>
> This is what I've tried so far:
> - Changing the number of modes. I tried 20, 100, 150 and 200.
> - Disabling boundary correction.
> - Disabling the intref option.
>
> None of these worked (although increasing the number of modes had a small positive impact). I would really appreciate if you could give me some advice to improve these segmentations.
>
> Thanks in advance,
> Miguel
>
>
