Outlining The Brain

The first stage towards measuring brain atrophy is to isolate the brain from the surrounding tissue, such as subcutaneous fat, eyes muscle etc. This stage can either be done automatically (with manual editing if necessary) using the "Find Brain" function, or manually by contouring the outline of the brain on each slice. In either case, the result will be a set of ROIs - one (or possibly more than one) in every slice that define the outer boundaries of the brain.

Load an image containing an MRI head scan covering the whole brain. The image contrast should be such that the CSF appears dark in the image. This could be:

  • Multi-slice T1-weighted.
  • Multi-slice FLAIR.
  • 3-dimensional acquisition T1-weighted.

You can also use the Brain Finder to isolate the brain on T2-weighted or Proton Density (PD) images, although you could not then go on to measure atrophy, since atrophy assessment only works with T1-weighted images.

Uniformity correction

It is strongly recommended that intensity uniformity correction is performed as part of the brain finding procedure, since this vastly improves the reproducibility. However, uniformity correction is quite time-consuming (a minute or so), and you may want to turn it off when not performing a detailed study, using the Checkbox to do uniformity (bias) correction check box. You should also set the Spinner to set the number of levels of detail for uniformity correction (See [the section on uniformity correction][unicorr.ms) for details). The default value of 3 should be sufficient.

There are two ways to isolate the brain.

1. Classic

Select the Classic tab:

The Classic Brain Finder method

Then set:

  • Threshold fraction. Brain Finder has one adjustable parameter that can be altered by the user to improve the resulting brain outline, the "threshold fraction" (default value 0.5). Brain Finder is pre-configured with a setting that works well for spin-echo T1-weighted MRI scans of the head. For 3-D T1-weighted and FLAIR scans, you should try values between 0.3 and 0.35. Clicking on the Set the tool defaults button puts the settings back to its default value.

    brain_finder_parameters

    The parameter acts as follows:

    • The surface of the brain is found at the intensity threshold between the interior of the brain and the background signal in the skull. A low setting of this threshold will push the brain outline further towards the skull, and a higher setting will pull the brain outline inwards. In FLAIR images, where the patient has a lot of hyperintensities, the mean intensity in the brain is much higher than at the cortical surface, and thus the intensity threshold will also be higher (for a fixed value of the threshold fraction).

    Note: whatever setting you use for your images, for both cross-sectional and serial studies it is important to maintain the same settings for all patients scanned on any given scanner. Because of differences in contrast from scanner to scanner, it may be necessary to adjust the setting on a scanner-by-scanner basis, but for serial studies it is imperative that the same settings are always used for any given patient.

  • Enhanced brain extraction. There may be errors in the brain outline that the Brain Finder tool finds, particularly around in the region of the eyes and around blood vessels. These tissues are often bright on T1-weighted images, particularly 2-D spin-echo images, and this option can help to remove some of the erroneous tissues.

    eyes

    Not enhanced

    eyes_enhanced

    Enhanced extraction

    Enhanced brain extraction works by identifying those tissue which are very much brighter (greater than 2.5 standard deviations from the mean) than the average brain intensity.

    Note: enhanced brain extraction adds somewhat to the time needed for brain extraction.

2. Mark 2

Select the Mark 2 tab:

The Brain Finder Mark 2 method

The Mark 2 method uses a template image with a pre-segmented brain mask to make brain segmentation more reliable. The template and brain mask are registered to your input image and used to initialise the segmentation, before it is iteratively refined to your image. You can either:

  • Use a built-in template, or
  • Supply a template image and brain mask.

The built-in templates are from human brains, whereas you can supply any template image you wish.

  • Using a built-in template

    Deselect the Use a built-in template checkbox. Then select the type of image contrast: T1-weighted (including FLAIR), T2-weighted, or proton-density (PD). The weighting selection makes sure that the correct template image is used.

  • Supplying a template and brain mask

    Select the Supplying a template image and brain mask checkbox. This will allow to select your own template image and brain mask:

    The Brain Finder Mark 2 method with a user-supplied template

    Set the template and brain mask images into the appropriate panels. Here are some hints for successful brain segmentation:

  • The template image should ideally have similar contrast to the image you want to segment, and must have the same orientation (axial/coronal/sagittal) and a similar field-of-view covering a similar range of anatomy.

  • The mask image must have the same dimensions as the template image (same number of rows/columns/slices in the image) and same pixel sizes. It is best produced by hand-segmenting the template image.
  • The mask image should have pixel intensity values between 0 and 1 (inclusive), with an intensity of 1 fully inside the brain and zero fully outside the brain. Any intensity values outside this range will be clipped.

Mark 2 does not require you to select a threshold fraction, and seems to be more reliable. However, registration of the template does take some time - up to several minutes, depending on the speed of your computer.

Normally, after registration of the template image, the brain outline is further iterated to make it conform more closely to brain brain surface. If you do not want this iteration, but just want the surface to be defined by the initial registered brain mask, then select Checkbox to prevent iteration after registering the brain mask image.

Writing the 3-D brain shape.

The Brain Finder will create ROIs surrounding the brain. However, you may want to visualise the brain surface in Jim's 3-dimensional display. If you select the Write the brain shape to a VRML file check box, then a surface model of the brain will be created, which can then be loaded into the 3-D Display as a virtual reality modelling language (VRML) file. The VRML file created will have a name taken from the input image name, with an extension .wrl.

Outlining the brain.

Next, click on the Button to find the brain outline button. This starts the uniformity correction (if selected) followed by an iterative search for the brain outline. During the search, a progress monitor pops up, which can be cancelled at any time:

brain_finder_progress

During the brain finding operation, the ROI toolkit is disabled.

When the brain has been found, ROIs are created that define the outline of the brain, and a message pops up:

brain_found_message

You now have a chance to review the brain outline found, and to edit the ROIs if you are not happy with Brain Finder's results. Click "OK" to acknowledge the message.

In some image slices, the brain may appear as multiple regions of interest. When editing the regions, you should retain all ROIs that actually contain brain parenchyma. In other slices, ROIs may have other ROIs within them, as in the example below:

brain_finder_hollow

This happens when the surface of the brain is concave; the internal ROI will not be counted as part of the brain.

You should now edit the ROIs if necessary, to correct the brain outlines found automatically. If you prefer, you can define the ROIs in every slice by manually outlining the brain, or by using ROI contouring. When you are happy with the ROIs in every slice, then proceed with calculating the BPF.

Note: you should save the ROIs as a permanent record of the brain outline found.