Print

Print


Hi,

Could you be more specific about how it is not "ideal"?
I'd really like to know whether there are some things which can be
improved, or whether it might depend more on differences between
where you want some borders to be placed and how the CMA
protocol places them (as FIRST is trained with data using the CMA
protocol).  To my eye the segmentation looked reasonable, and
certainly not out by a factor of 4, which is what your original results
were saying.  Consequently I really do need your help to isolate any
potential problems.  If you could please send back some images or
coordinates then that would be the most helpful.

As for the Morey paper - it was run with a much earlier version of
FIRST and used a very simplistic method of boundary correction,
so it isn't very relevant anymore.

All the best,
	Mark


On 28 Oct 2009, at 00:00, Jay Ives wrote:

> Hi Mark,
>
> Thanks for the advice. Indeed this was the problem case. Processing  
> with preliminary cropping makes a small difference (less than 2%).
>
> I don't think the hippocampus segmentation is ideal, however it is  
> interesting that FIRST demonstrates a 9% size difference between  
> right and left which agrees with my clinical impression, whereas  
> FreeSurfer measures both sides to be the same.
>
> I read the paper "A comparison of automated segmentation and manual  
> tracing for quantifying hippocampal and amygdala volumes
> by Rajendra A. Morey et al. NeuroImage 45 (2009) 855-866"
>
>
>
> Do you have any comments on this?
>
>
>
> Cheers.................J
>
>
> ----- Original Message ----- From: "Mark Jenkinson" <[log in to unmask] 
> >
> To: <[log in to unmask]>
> Sent: Tuesday, October 27, 2009 10:56 PM
> Subject: Re: [FSL] First and hippocampus
>
>
>> Hi,
>>
>> Your images have a reasonable amount of neck in them which makes
>> registration difficult.  The best thing is to crop the images to   
>> remove the
>> neck and then reorient to axial in order to make the registration  
>> work
>> the best.  If you do the crop and reorient and then do  
>> run_first_all  then
>> you should get a good segmentation.  At least it looks good to me,  
>> although
>> I'm not an anatomical expert.  Try the following and see what you  
>> think
>> of the results:
>>  fslroi jay_ives_data.nii head 0 512 145 335 0 175
>>  fslswapdim head.nii.gz RL PA IS head_axial
>>  run_first_all -i head_axial.nii.gz -o head_axial_seg
>> and look at head_axial_seg_all_fast_firstseg overlayed on the
>> head_axial image.  If you see things there which you think are
>> not good then please let me know - ideally with some images
>> or coordinates where you can point out the problems.
>>
>> If you are happy with the above then I suggest that you go
>> through this process each time.  The first stage needs to be
>> done by hand, determining the right place to crop the image
>> using fslview - but it doesn't take very long to do - at most a
>> minute per image, which isn't that onerous.
>>
>> By the way, I'm assuming that this was the difficult/problematic case
>> from before - right?  If not, then I really need to see the  
>> problematic
>> case.
>>
>> All the best,
>> Mark
>>
>>
>> On 27 Oct 2009, at 12:30, Jay Ives wrote:
>>
>>> Hi Mark,
>>>
>>> I have uploaded a file jay_ives_data.nii to your site. The  
>>> reference number is 398602.
>>>
>>> Thanks
>>>
>>> J
>>>
>>> ----- Original Message ----- From: "Mark Jenkinson" <[log in to unmask]
>>> >
>>> To: <[log in to unmask]>
>>> Sent: Tuesday, October 27, 2009 4:55 PM
>>> Subject: Re: [FSL] First and hippocampus
>>>
>>>
>>>> Hi,
>>>>
>>>> I suspect the problem is the initial registration if you find the  
>>>> results change
>>>> when you reslice to axial.  It will only affect the ability of  
>>>> the model (which is
>>>> in standard space) to be transferred to the native image.  Once   
>>>> the model
>>>> is in the native image space it makes no difference what   
>>>> orientation it is in.
>>>>
>>>> As for trying to improve your fit - I'm afraid there really is   
>>>> nothing I can
>>>> do based on a worded description.  I really need to see the  
>>>> data.   Can you
>>>> please upload some data so that we can see?
>>>>
>>>> All the best,
>>>> Mark
>>>>
>>>>
>>>> On 26 Oct 2009, at 23:38, Jay Ives wrote:
>>>>
>>>>> Hi Mark,
>>>>>
>>>>> Well, I ran it again after doing 'fslswapdim RL PA IS' to get an  
>>>>> axial format, and the results were much improved and overlay    
>>>>> properly on the axial format data. I'm surprised I had to do   
>>>>> this,  but it's no big problem.
>>>>>
>>>>> I wrote to Brian Patenaude last July about the less-than-ideal   
>>>>> fit  of the model to the hippocampus. Unfortunately, this is  
>>>>> still  an  issue despite the new version. I have run 'run_first'  
>>>>> with  several different modes (30 to 60) and with intensity  
>>>>> referencing  to the thalamus, but could not get a satisfactory  
>>>>> fit. This is  most noticable at the pes hippocampus and at the  
>>>>> lateral  hippocampus body.
>>>>>
>>>>> It is a pity beacause it would be nice to use the tool in   
>>>>> analysis, but it just isn't accurately defining the structure  
>>>>> of  interest. If anything can be done, please let me know.
>>>>>
>>>>> Thx........J
>>>>>
>>>>> ----- Original Message ----- From: "Mark Jenkinson" <[log in to unmask]
>>>>> >
>>>>> To: <[log in to unmask]>
>>>>> Sent: Monday, October 26, 2009 9:57 PM
>>>>> Subject: Re: [FSL] First and hippocampus
>>>>>
>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> This is odd.
>>>>>> The image that is fed into run_first_all as the input and
>>>>>> the output segmentations of run_first_all should have
>>>>>> exactly the same size.  They will be different from the
>>>>>> FreeSurfer inputs and outputs normally, but there should
>>>>>> be no inconsistency between the input and output to
>>>>>> run_first_all.
>>>>>>
>>>>>> All the best,
>>>>>> Mark
>>>>>>
>>>>>>
>>>>>> On 26 Oct 2009, at 11:56, Jay Ives wrote:
>>>>>>
>>>>>>> Hi Mark,
>>>>>>>
>>>>>>> OK, that's fine...but can one overlay the segmentations on the  
>>>>>>> original images to see how good the fit is?
>>>>>>> I tried this but the orientation of the segmentations is quite  
>>>>>>> different from the original data. Why?
>>>>>>> Should I have fed axial data into FIRST?
>>>>>>>
>>>>>>> Thanks......J
>>>>>>>
>>>>>>> ----- Original Message ----- From: "Mark Jenkinson" <[log in to unmask]
>>>>>>> >
>>>>>>> To: <[log in to unmask]>
>>>>>>> Sent: Monday, October 26, 2009 5:27 PM
>>>>>>> Subject: Re: [FSL] First and hippocampus
>>>>>>>
>>>>>>>
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> To look at the first segmentations you simply load the image
>>>>>>>> subjectname_all_fast_firstseg.nii.gz into FSLView and look at  
>>>>>>>> it.
>>>>>>>> You may need to set the colormap to the MGH-Subcortical
>>>>>>>> one (although this may be the default).
>>>>>>>>
>>>>>>>> If it looks like the segmentations are doing badly, then please
>>>>>>>> upload the data to our upload site:
>>>>>>>> http://www.fmrib.ox.ac.uk/cgi-bin/upload.cgi
>>>>>>>> and send us the reference number.  We'll then have a look
>>>>>>>> and see if there's anything that can be done to improve them.
>>>>>>>>
>>>>>>>> All the best,
>>>>>>>> Mark
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> On 25 Oct 2009, at 12:00, Jay Ives wrote:
>>>>>>>>
>>>>>>>>> Thanks Mark,
>>>>>>>>>
>>>>>>>>> The FreeSurfer segmentations look pretty good - perhaps a   
>>>>>>>>> little over-generous.
>>>>>>>>> I actually am having trouble working out how to visualise   
>>>>>>>>> the FIRST segmentation output - can you please advise me?
>>>>>>>>>
>>>>>>>>> Yes, I am running FSL 4.1.4
>>>>>>>>>
>>>>>>>>> I got the volumes from FIRST using the commands:
>>>>>>>>>
>>>>>>>>> fslstats subjectname_all_fast_firstseg.nii.gz -l 16 -u 18 -V
>>>>>>>>> fslstats subjectname_all_fast_firstseg.nii.gz -l 52 -u 54 -V
>>>>>>>>>
>>>>>>>>> The FreeSurfer volumes are in aseg.stats - as I'm sure you  
>>>>>>>>> know.
>>>>>>>>>
>>>>>>>>> Cheers.......J
>>>>>>>>>
>>>>>>>>> ----- Original Message ----- From: "Mark Jenkinson" <[log in to unmask]
>>>>>>>>> >
>>>>>>>>> To: <[log in to unmask]>
>>>>>>>>> Sent: Sunday, October 25, 2009 7:15 PM
>>>>>>>>> Subject: Re: [FSL] First and hippocampus
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>> Hi,
>>>>>>>>>>
>>>>>>>>>> Firstly, did you look at the output images?
>>>>>>>>>> This is the best way to tell if things are going wrong or  
>>>>>>>>>> not.
>>>>>>>>>> It should give you an indication of whether there is some
>>>>>>>>>> systematic over/under-estimation.  You should look at
>>>>>>>>>> both the FIRST output segmentations and the FreeSurfer
>>>>>>>>>> output segmentations.
>>>>>>>>>>
>>>>>>>>>> Secondly, are you running the FSL4.1.4 version?  There
>>>>>>>>>> are significant differences which are worth making sure
>>>>>>>>>> that you are running.
>>>>>>>>>>
>>>>>>>>>> Thirdly, what commands are you using to get the volume
>>>>>>>>>> measurements from the segmentations?
>>>>>>>>>>
>>>>>>>>>> All the best,
>>>>>>>>>> Mark
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On 25 Oct 2009, at 09:49, Jay Ives wrote:
>>>>>>>>>>
>>>>>>>>>>> Hmmm.....I've run FIRST and FreeSurfer on the same data  
>>>>>>>>>>> for  2 subjects and
>>>>>>>>>>> got quite disparate results for hippocampus volumes. In    
>>>>>>>>>>> fact, they are not
>>>>>>>>>>> even close.
>>>>>>>>>>>
>>>>>>>>>>> Subject 1:  L 1317 (FIRST)  5068 (FreeSurfer)
>>>>>>>>>>>           R 1855             5054
>>>>>>>>>>>
>>>>>>>>>>> Subject 2:  L 2916             4934
>>>>>>>>>>>           R 4300             5325
>>>>>>>>>>>
>>>>>>>>>>> My data is good quality T1 volumes from a 3T Verio system
>>>>>>>>>>>
>>>>>>>>>>> FIRST version is 1.2 (command run_first_all)
>>>>>>>>>>> FS version is 4.4.0   (command recon-all -s subject -all)
>>>>>>>>>>>
>>>>>>>>>>> So no fancy stuff. Any comments?
>>>>>>>>>>>
>>>>>>>>>>> Thx............J
>>>>>>>>>>>
>>>>>>>>>
>>>>>>>
>>>>>
>>>
>