SpinozaWiki:Brain Extraction Tool

Link to BET webpage [1]

Brain extraction: purpose

Apart from the fMRI scans, a MRI scan is usually also recorded. Recording the MRI scan serves three purposes.

(1) A MRI scan improves the process of normalizing the fMRI scan to a standard space. This step is important because you want to report, in the end, on the activation at the group level (referring to the entire group). The brain of individual subjects differ substantially from each other and therefore the fMRI scans need to be normalized towards a standard space. This will be the topic of week 4.

(2) The second purpose comes into play when activation of different groups is compared. Lets say you have a group of old and young subjects and you want to compare activations measured on a particular task between these groups. It could be that the measured activation differs between these groups because the brains of these subjects function in a different way. However, it is also possible that the activation between the groups differ from because their brains differ structurally from each other. For this you use voxel based morphometry after brain extraction.

(3) You might want to see to which degree gray matter differences covary with a second measure, for instance intelligence or aging. For this you also use voxel based morphometry after brain extraction.

Brain extraction: process

Brain extraction is the process of removing the skull and the rest of the head from the brain and it can be done with the FSL program Bet. The resulting file only contains a representation of the brain's anatomy and this makes it possible to align the fMRI scan to the MRI scan of that subject. This file is saved with a ‘_brain’ extension at its end. After aligning the scans of a single subject, the next step is to align the MRI brain scans of different individuals with a representation of a standard brain (based on 152 subjects). You can start BET from the console (with the command bet). Do not use the BET-GUI because we almost always need options that are not available in the GUI. Also always use FSLView to inspect the output of the bet procedure and to fine-tune the bet operation. Run bet in the console to see the available options and its command line:

bet <input file> <output file>  <option 1>  <value option 1>  <option 2>  <value option 2> etc. 

(don't take the '<,>' literally!)

You can use the [tab] option when you enter the filenames for correct and fast completion of the filenames. Give the output file the same name as the input file but change the nii.gz extension to '_brain'. This is important for the batch process at a later stage. Use the add file option in fslview to view the result of the bet procedure. Use it to lay the betted brain over the original image. Change the colour of the betted file to red. This will make it easier to differentiate the extracted brain image from the original brain image. Often the results of the first try are not very good. In general the estimation of the brain outline is too small. We can improve the results of the bet procedure with the options -c, -f and –g (see sections below).

Setting the centre of the brain

With option -c you can indicate where the centre of the brain is located. BET assumes that the centre of the brain is located in the centre of the image. This assumption is not always correct and depends on how the brain is recorded. Note that values should be entered in voxels, not mm. Using the -R command will let BET automatically determine the centre of the brain by recursively adjusting it to the new centre of the extracted brain for a maximum of 10 iterations.

Estimating the outline of the brain

With the option -f you can indicate whether BET should include a larger or smaller area in the estimation of the brain outline. The default value of this estimate is 0.5. A smaller value gives a larger brain outline. A larger value gives a smaller brain outline.

Estimating the vertical length of the brain

With the option -g you influence the vertical gradient of the estimated brain size. So with this option you can indicate whether more of the top or more of the bottom of the selected area should be included in the estimation of the brain. The default value is 0. Larger values give a larger brain outline at the bottom. Smaller (negative) values give a larger brain outline at the top.

Example

Sets the centre of the estimation at 125,81,130 mm, generates a smaller brain outline then default (the 0.6), and generates a larger brain outline at the top.

bet ppSC.nii.gz ppSC_bet -c 125 81 130 -f 0.6 -g -0.5

Automating BET

BET extraction can be run automatically on a large dataset with the function BetCrawler (see BetCrawler). When this function is run it is necessary to check the slicesdir subdirectory that will also be created. Brains that are not extracted properly need to be extracted manually with the –c option.

Generating a brain mask

When you perform calculation with an image a program will not know beforehand where relevant signals are located (for instance which parts of the image belong to the brain and which parts belong to the skull and the rest of the head). If the program knows the location of the brain in the picture, calculations can be performed on relevant information. One way of speeding up calculations is to remove non relevant parts from the fMRI images (see the section on ‘Data conversion & removing of non-relevant parts from the fMRI scans’). Another way is to explicitly estimate where the brain is located. You can generate a so called ‘brain-mask’ by adding the option –m to the bet command line. This mask will make sure that only voxels within the mask-area are used for further analysis. Adding the option –m to the bet command will generate such a mask.

bet pp0079-20071020-0003.nii.gz pp0079-20071020-0003_brain -c 125 81 130 -f 0.6 -g -0.5 –m