Look at https://github.com/mjirik/lisa. It has been developed mainly for liver segmentation but it can be used also for other body parts. It is based on the graph cut method, it is written in Python + Numpy, Scipy and supports Matlab import/export.

If you have the T1 or T2 images in Nifti format you can use the segmentation functions from the SPM package (http://www.fil.ion.ucl.ac.uk/spm/). By using those functions you can construct the gray and white matter maps, load them both in Matlab (also using the SPM functions) and everything should be straigtforward.

I could also recommend you to use IBASPM, which is a SPM-based toolbox, also available from the SPM webpage.

Hi William,

If you're already using MRtrix, you might want to check out the command "dwi2mask"; it's a very simple strategy to get a whole brain mask (including CSF) from your diffusion data, but it's quite robust in it's simplicity as well.  Otherwise, I'd definitely recommend to get further into FreeSurfer for that specific task as well.

@Thijs, I did initially think of that, but when I played around with it, I couldn't extract a brain from a T1 image in FSL or MRTrix, I have got FreeSurfer working now, so I will play around with that, it seems to be the most well documented one.  My understanding was that that command was for DWI images only, I was guessing the segmentation worked on the boundaries of the diffusion tensor?

@Sabine thanks, if I cannot get thngs working in FreeSurfer in the next couple of days I will have another look at Slicer, it's a matter of time and learning to use just a couple of tools really well for me, otherwise I spend weeks installing software and playing about with it without clear direction. 

Thanks everyone for your answers. 

Yes, dwi2mask (as the name implies ;-)) is only for DWI images, not for any other contrast.  What it essentially does is:

1) Get the average DWI for each distinct b-value in the data (including for b=0).

2) Find an optimal threshold for each of those b-value images.

3) Just apply those thresholds and merge the masks together.

In practice, you'll find that an average DWI for anything with a significant b-value (that includes "low" b-values such as 700s/mm^2 or so) by itself already more or less nicely separates the brain from its immediate surroundings.  Only the CSF has low (to very low, essentially almost zero) intensities for higher b-values.  But the b=0 more or less separates the CSF from the rest.  So combining those masks gives you an (admittedly rough) brain mask.  Advantages are that it's directly derived from the DWI dataset.  Well, advantage being that that comes in handy if you need it just in relation to the DWI data of course (e.g. as a tracking mask, or a mask to do any other DWI computations in, like CSD, etc...).  So in those scenarios, you're not dependent on a coregistered T1 (or other anatomical contrasted) image.

If you're after an accurate brain mask for other applications (let's say brain volume measurements or the like), it's definitely not advised to use that command.  For anything else than essentially a DWI-based brain mask, I'd first look at FreeSurfer.

Hi William,

If you have FreeSurfer working, I'm sure it will do what you want. Since you also have FSL installed, you can use "bet" to do brain extraction and use "fast" to segment the brain into grey matter, white matter, and csf. There are both guis and command line options for these that are very easy to use. I have included links to both.

http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/BET/UserGuide

http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FAST

Ah, thanks Avery, I did try BET but it didn't work and I assumed that was because I was using a T1 scan. The other thing is that I am using mouse imaging data, that shouldn't make a difference though, should it? 

Hmm, most of the tools mentioned (FSL, FreeSurfer, SPM) are optimized for human brain segmentation from T1-weighted images. You could look into SPM with the SPMMouse extension though.

http://www.spmmouse.org/

Yeah... well... :-)  I personally wouldn't really advise to use BET in any way.  In my experience, it's everything but robust in a multitude of scenarios.  It's ok to get maybe an initial brain mask-y output, that you can further edit manually (as, e.g., in: fill in the regions where BET happily cut into the brain); but for quantitative purposes... don't trust it.  And that's even for human brain data.  The ad hoc assumptions that it makes, are definitely not suited for mouse data.

Darn it all, I was hoping that it was not making too many assumptions and just running some kind of Monte-Carlo simulation to segment purely on numerical grounds, the scalar density of the image value or something.  I have been working on it this morning, FSL gives a pretty dodgy-looking set of segments and that's without looking at the quantitative data. 

Basically I want to segment out the brain so I can look at the region of some glioblastomas we are simulating in the mouse model,  then I will need to extract this segmentation to MATLAB to get a gradient at the tumour boundary for T1 vs T2 data to gie an estimate of the boundary which we can then compare to the isosurfaces in our virtual model.  Yes, I know there is a heap of dodgy stuff there in that we really have no idea how in vivo imaging correlates quantitatively to cell density, but it's not really amenable to experimental validation, so we have to make some assumptions somewhere. 

My plan was to segment the brain, essentially even just skull stripping would do the trick, then export this data into MATLAB so I can have a good sift through it. 

I'll look at SPM, the others are not really working for me.  Thanks for all your input so far, very useful :)

FSL BET toolkit is good for skull stripping and brain segmentation.

But if you are interested in regional brain segmentation then you can try ITKSNAP.

My colleague here at who gave me the mouse data says he uses ITK-SNAP, which I have, so I will try that and SPM with the mouse tool kit.  For Humans, FSL works well, but it won't process the mouse data.

Dear William,

Because we were dissatisfied with existing brain ± intracranial volume extraction techniques (cf. Thijs' comment yesterday), we developed a new one ("pincram") based on multi-atlas methodology.

Crucially, it works with example data from _other_ scanners. Therefore, you should be able to use existing mouse brain atlases.

We've recently published it open-access (see attached), and the paper contains the link to the software. Besides the flexibiltiy in using atlases / examples from anywhere, it also outperformed virtually all other methods (see paper).

Hope this helps,

All the best,

Alexander

-----------------------------------------

Alexander Hammers, MD PhD

Professor (Honorary Consultant) of Imaging and Neuroscience

Head of PET Imaging Centre

Division of Imaging Sciences and Biomedical Engineering

King's College London

St Thomas' Hospital, London

Telephone +44-(0)20 7188 8364 (PA Viviane Schroeder)

Email [email protected]

Article Brain Extraction Using Label Propagation and Group Agreement: Pincram

Excellent, thank you so much Alexander, I will check it out.  I did a bit of what I want with ITK-SNAP today, but I had to cut out two different shaped rectangular boxes to contain the tumour I want to look at in MATLAB for the T1 and T2 images, if I could strip the skull I could co-register the images and superpose the brains and this would make what I want to do a lot easier.

Dear William,

Thanks. I should probably have mentioned that we worked (and validated) on humans not mice. However, the generalisability is there; the most important question to check whether it would work with your data is whether the registration steps work.

I am reasonably confident as we've previously adapted an (anatomical) multi-atlas brain segmentation method, MAPER (Heckemann RA et al. Neuroimage 2010), for (rat) brain segmentation (Yankam Njiwa et al. submitted). You also seem to have working tools for registering mouse brains already.

If you run into difficulties we can check whether we can help - always interesting to extend tools to new uses!

All the best,

Alexander

kindly provide some software for brain image segmentation before we apply algorithms