surface reconstruction

Questions about BrainVisa usage and installation

Moderators: denghien, riviere

User avatar
Melanie PELEGRINI
Posts: 39
Joined: Tue Mar 23, 2004 4:20 pm
Location: U678 Inserm, Paris

surface reconstruction

Post by Melanie PELEGRINI »

Dear all
Below is a message from a colleague who does not have his subscription to the forum yet ;-)
Melanie



======================================================
Dear BrainVisa Experts,

I am beginner of BrainVisa, and I have a question about surface reconstruction. I did surface reconstruction using T1 pipeline default values, and I found part of surface is outside of the brain (see attached figure). I think this problem results from gray/white matter segmentation in the preprocessing step. Does anyone know how to solve this problem?

Thanks in advance
Xingfeng
Screenshot-7.jpg
Screenshot-7.jpg (150.75 KiB) Viewed 11716 times
User avatar
sunboy4554
Posts: 18
Joined: Thu Jul 31, 2008 1:38 pm
Location: China

Re: surface reconstruction

Post by sunboy4554 »

Hi
I want to know do you have a perfect brain mask and split mask? If so, is your data good enough? You can show us the MR scan parameters, and we will find where the problem are.

sunboy
[img]Brain%20is%20too%20hard%20to%20Know%20and%20master%20it.[/img]
User avatar
Jean-Francois Mangin
Posts: 337
Joined: Mon Mar 01, 2004 10:24 am
Location: Neurospin, CEA, France
Contact:

Re: surface reconstruction

Post by Jean-Francois Mangin »

Hi,
if your dataset is not working with the default parameters, this could come from plenty
of reasons. You should go step by step, following various issues in the forum.
Considering the ugly data you show, any guess is bound to be wrong...
Xingfeng Li
Posts: 8
Joined: Wed Dec 17, 2008 10:17 am

Re: surface reconstruction

Post by Xingfeng Li »

Hello,
I am doing this work now. What i have done is as follows:

First, I used 'Import T1 MRI' to load the data
Second, i run 'T1 Pipeline 2007'
I selected Normalised: MNI from mritotal, because i have transferred all the data to MNI space using mritotal.
l leave other parameters as it is (default)
in 'T1 Pipeline 2007' window, I selected all steps, including T1 bias correction, Split Brain Mask, Grey white interface, Cortical fold graph etc.

then RUN the program,
it gave the reuslts as attached figure.

I am going to redo the surface reconstruction. What i should do, should i do it step by step? I should do one step and check one step, and then going on. It could be painful to do for several hundred subjects. Any good idea?

Thanks a lot.
Xingfeng Li
Posts: 8
Joined: Wed Dec 17, 2008 10:17 am

Re: surface reconstruction

Post by Xingfeng Li »

Hi,
I also have some addation informations.

I repeated the calculation for three subjects, results of two subjects are fine, but the the third one is not good. During the calculation, it gives:




Process T1 Pipeline 2007 134 started on 2008/12/18 15:43
------------------------------
! VIP Warning:
Something was wrong with the iterative erosion!
------------------------------
Computing left hemisphere grey-white classification...
Computing right hemisphere grey-white classification...
New cubic spatial resolution:1
Computing oversampled MR image to improve cortical surface definition (cubic spline)
Computing oversampled voronoi diagram (nearest neighbor)
Masking Bias corrected image with left hemisphere mask...
Detecting left grey/white interface...
Reconstructing left hemisphere white surface...
Triangulation and Decimation...
Warning : null normal encountered
Warning : null normal encountered
Warning : null normal encountered
Warning : null normal encountered
Smoothing mesh...
Masking Bias corrected image with right hemisphere mask...
Detecting right grey/white interface...
Reconstructing right hemisphere white surface...
Triangulation and Decimation...
Warning : null normal encountered
Warning : null normal encountered
Warning : null normal encountered
Smoothing mesh...
Masking Bias corrected image with left hemisphere mask...
Detecting left grey/white interface...
Reconstructing left hemisphere surface...
Triangulation and Decimation...
Masking Bias corrected image with right hemisphere mask...
Detecting right grey/white interface...
Reconstructing right hemisphere surface...
Triangulation and Decimation...
Masking Bias corrected image with hemisphere masks...
Detecting left grey/white interface...
Computing skeleton and buried gyrus watershed...
Building Attributed Relational Graph...
computing additional attributes
Detecting right grey/white interface...
Computing skeleton and buried gyrus watershed...
Building Attributed Relational Graph...
computing additional attributes
Skip unselected node: SulciRecognition

Process T1 Pipeline 2007 134 finished on 2008/12/18 15:58 (14 minutes 47 seconds)





I attached the surface of this third subject.

Thanks.
Attachments
test.jpg
test.jpg (147.82 KiB) Viewed 11672 times
User avatar
sunboy4554
Posts: 18
Joined: Thu Jul 31, 2008 1:38 pm
Location: China

Re: surface reconstruction

Post by sunboy4554 »

Hi
Let me know your data parameters (TR/TE). Maybe there is something wrong with the data.

Sunboy
[img]Brain%20is%20too%20hard%20to%20Know%20and%20master%20it.[/img]
Xingfeng Li
Posts: 8
Joined: Wed Dec 17, 2008 10:17 am

Re: surface reconstruction

Post by Xingfeng Li »

Thanks a lot for replying.
My MRI data acquistion details are:

sequence = MP-RAGE
TR(msec) = 9.7
TE(msec) = 4.0
flip angle(degree) = 10
T1(msec) = 20
TD(msec) = 200
orientation = sagittal

All data were resampled to the Tal space by using mritotal before reconstruction. All my data are MINC format.
User avatar
sunboy4554
Posts: 18
Joined: Thu Jul 31, 2008 1:38 pm
Location: China

Re: surface reconstruction

Post by sunboy4554 »

Hi
As I know, the TR time may be a little shorter, which may result in a poor management on your data. And why didn't you use the BV pipeline? There may be some unexpected problems during the software changing.
Sunboy
[img]Brain%20is%20too%20hard%20to%20Know%20and%20master%20it.[/img]
User avatar
Jean-Francois Mangin
Posts: 337
Joined: Mon Mar 01, 2004 10:24 am
Location: Neurospin, CEA, France
Contact:

Re: surface reconstruction

Post by Jean-Francois Mangin »

If you have two subjects working, proceed further with your database and keep the ones working. Once again anything can happen with the third, and you do not give any clue
(you do not realize your messages are just the highest level ones, you should open the log window to get more, I suspect yur
histogram analysis is wrong).

And relative to processing hundred of subjects, yes you should tune the parameters performing each step one by one on a sample of the dataset.
It is not what you did. Anyway, without some help by somebody with experience, or taking more time than you did reading documentations,
you can not process hundred of data. The documentation and the forum are supposed to be read. It is impossible to do individual
teaching from internet for each user, sorry...
Xingfeng Li
Posts: 8
Joined: Wed Dec 17, 2008 10:17 am

Re: surface reconstruction

Post by Xingfeng Li »

Thanks a lot.

I have an idea to handle this problem,i.e. using FSL (fslvbm_1_bet to extract brain information) to cut the skull, then run BrainVisa after that. This is based on the fact that BrainVisa reconstructs large parts of skull for the whole brain, if the skull can be removed beforehand, the cortex surface could be fine, at least for my data.

Anyway, i'd like to suggest to improve the segmentation of BrainVisa, just using FSL's BET to cut the skull before do further. This is not too difficult to do, to my knowledge.

I have not tried this method, because i have troubles to transfer data to CENIR, i will verify this idea next year.
User avatar
riviere
Site Admin
Posts: 1361
Joined: Tue Jan 06, 2004 12:21 pm
Location: CEA NeuroSpin, Saint Aubin, France
Contact:

Re: surface reconstruction

Post by riviere »

Hi,
In most cases it is just a matter of changing a parameter in the bias correction step, but as Jeff states, we cannot help you without more knowledge of your problem.
But in any way, if using FSL as a first step helps to make the pipeline more robust... why not... Your idea looks a little bit like what Jeff does to process monkey brains, so... it might work... but on the other hand if the problem is really due to a bad bias correction, you may still get a bad grey/white segmentation.
Denis
User avatar
Jean-Francois Mangin
Posts: 337
Joined: Mon Mar 01, 2004 10:24 am
Location: Neurospin, CEA, France
Contact:

Re: surface reconstruction

Post by Jean-Francois Mangin »

Peter Kochunov usually applies FSL to get the brain mask, for humans and monkeys (maybe you can get information on his web page).
I never used FSL myself,
I have no feeling about the behaviour of BET. The story is not just to get a brain mask. And I do not think it is the brain mask
extraction that is problematic with your data, maybe the histogram analysis (bias correction is robust nowadays).
There is a bug in this command appearing sometime. A usual turnaround is to add -u 16, or -u 8 to the command you get
in the log window.
I think if you want to use BET, you have to put yourself the mask at the right place, and also find a way
to profide the histogram analysis (this is a very simple ASCII file with han extension). This file is used further
in the pipelines in several places. If you have a grey/white classification of the mask, you can find a Vip command creating this file I think.
If you are coureageous enough to do a brainVISA pipeline mixing BET with our tools, please give it back to the community.
If it is not a brainVISA pipeline, at least provide the way you made it work to others, this is the goal of the forum.

In fact we have no problem with alternative pipelines using tools from other packages. We even do it ourselves
but we can not do everything. BrainVISA, which is a huge burden for our team, has been developed just for that.
We will never claim we have the best algorithm everywhere, it would be ridiculous. You can not figure out the incredible
variability of the MR images we get in touch with. The default parameters have been set because we need default,
but experienced users should try some tuning. Once you found a method working with your data with
one package (BrainVISA, FSL or other), it is usually a good idea to use it...

To be honest, the weakest part of our pipeline is the method splitting the brain into pieces. With some
MR setting, you get bad hemisphere/cerebellum interface whatever the tuning, and you need some hand-based correction
if you are interested in this area.

And to conclude, I have processed myself more than 2000 brains with brainVISA, including humans of any ages, baboons, macaques, chimps...
The way I do it could be surprising to people familiar with VBM, which is usually triggered as a black box. I check visually each slice
of each intermediate result (brain mask, brain split) and a few other things... Hence I perform corrections when required and discard the data if it is
hopeless (it is rare if you are experienced). I do not want to spend months trying to interpret
some morphometric result if some data are just crap. You should realize that extracting and recognizing the sulci is more risky
that doing VBM. Hence you need a good brain mask not full of surrounding tissues...
A lot of things can fail otherwise. If you are interested in your data, you can invest a few days into the processing.
I would say that with data of reasonable quality, processing 100 brains in a week checking everything is possible
for an experienced user.
Xingfeng Li
Posts: 8
Joined: Wed Dec 17, 2008 10:17 am

Re: surface reconstruction

Post by Xingfeng Li »

Hi Jean-François Mangin,

Yes, you are right. There are other problems when i used skull removed brain to do the reconstruction. I should add the skull removed brain to some step, at least not at the beginning of the processing. But i do not know where i should add.

Is it possible to visit your lab to show me how to do the reconstruction? all my data are MNC format and in Tal space.

Thanks a lot.
User avatar
Jean-Francois Mangin
Posts: 337
Joined: Mon Mar 01, 2004 10:24 am
Location: Neurospin, CEA, France
Contact:

Re: surface reconstruction

Post by Jean-Francois Mangin »

The skull removed brain should be turned into a mask and replace the brain_mask generated by brainvisa.
You work in Paris? Yes you could visit us.
Xingfeng Li
Posts: 8
Joined: Wed Dec 17, 2008 10:17 am

Re: surface reconstruction

Post by Xingfeng Li »

Hi Jean-Francois Mangin,
Thanks a lot.
I have to say i am a beginner. After I have tried several times, i found it works! The results are much better! In addition, it is also faster if i cut the skull before do anything else.

I am attaching 2 figures for comparison. The first one is the original brain( without cut skull) , the second one is the reconstruction result after i remove the brain skull and do the same thing.

I would suggest to cut the brain skull ( for example using FSL BET) and then do the reconstruction using BrainVisa.

All the best
Attachments
remove brain skull
remove brain skull
cut_skull_surface.jpg (155.28 KiB) Viewed 11534 times
not remove brain skull
not remove brain skull
not_cut_skull.jpg (146.4 KiB) Viewed 11532 times
Post Reply