Afni Masks

SPM Masks

This webpage is not a substitute for that tutorial, nor is that tutorial a substitute for this page. (See the Afni Masks link above.)

Afni offers specialized tools for creating masks on high resolution anatomicals. You can

1. Freehand draw a mask (see below) or,
2. If you use talairached images, you can select a brain region or group of brain regions (see below) to be included in the mask (the above tutorial describes these functions in detail and provides helpful screenshots).

The tool 3dfractionize can be used to convert these high resolution masks to lower resolution. The conversion is necessary if you intend to multiply a lower resolution image by the mask.

3dAutomask automatically creates a mask of brain from a 3D+time image. The resulting image is a 3D fim (functional) image with 1's in the brain and 0's outside. Such an image can be overlayed on an anatomical to see signal dropoff or can be used like other masks to filter an image.

3dcalc can be used to multiply an image of interest by the mask you create.

Mask Multiplication
Each slice of an image is a matrix (square or rectangle) of numbers...0=black (no activation) and other numbers represent levels of grey. A mask is usually binary: it only has 1's and 0's in it, and 0=black and 1=white. To multiply two images, you dot-multiply the two matrices, which means that the top right number in one matrix is multiplied by the top right number in the other matrix, the second in one matrix is multiplied by the second in the other matrix, etc. To multiply two image matrices, those matrices should be the same size--i.e., same number of slices and same number of voxels in each direction. If you are multiplying an image by a binary mask, then any value in the original image is retained if it is multiplied by 1 but zeroed out if multiplied by 0. e.g.,

In our example, 2 4x4 matrices are multiplied. The binary mask matrix serves to filter out any values in the activation image that occur outside the inner rectangle.

Afni Tutorial Data
You can use data from the afni tutorials to play with masking. Create a directory to work in and put all these files in the directory. To really appreciate what talairaching does for you, you should initially dump the tlrc files in a separate subdirectory (If you don't afni will find the talairached files and allow you to do all talairach related options). Go here for:

• 2D anatomicals: brain+orig.BRIK and HEAD and brain+tlrc.BRIK and HEAD
• 3D anatomicals: brain3d+orig.BRIK and HEAD and brain3d+tlrc.BRIK and HEAD
  
• An analyzed functional image (a 3D fico image): study1_block80_norm_p05_150ul+orig.BRIK and HEAD
  

Go here for:

• A functional 3D+time image (this is the result of preprocessing but before analysis): study1_norm+orig.BRIK and HEAD

Data:

• brain+orig.BRIK and HEAD to draw a mask
• study1_block80_norm_p05_150ul+orig.BRIK and HEAD as a template file for 3dfractionize
• study1_norm+orig.BRIK and HEAD to multiply by your mask
  

Procedure:

• Open afni and choose Define Datamode->Plugins->Draw Dataset.
• In the Draw Dataset interface, select the brain+orig file to draw on.
• Use the middle mouse button to draw and fill shapes. (If you intend to convert it to low resolution you will need big bold shapes and/or you may need to play with clipping in the 3dfractionize command). For example, you might draw a good sized filled box in the upper left corner of slice 5.
• Use "save as" to give the result file a new name (e.g., mask).
• Insides of filled shapes are 1's by default. Everything else is 0's in your mask. (If you want more options, read the tutorial mentioned above).

(e.g., 64x64xNAS)
You can create a courser grid file from your above ROI drawing:

>3dfractionize -template study1_block80_norm_p05_150ul+orig -input mask+orig -clip 0.5 -preserve -prefix outfimfunc

Pick a functional template from the same run, use your mask as input, preserve the original data inside the mask and designate a prefix for your output. The output (e.g., outfimfunc+orig) is a sea of zeros except for your designated mask voxels, all nicely aligned to their new 4D parent. The output will be 3d low resolution.

By using a fim file (post-analysis) as your template, your output will be a fim functional file which you can overlay on an anatomical.

To learn more, you might also type:

>3dinfo out*.BRIK

>3dcalc -prefix calc_out -a study1_norm+orig -b outfimfunc+orig -expr 'a*b'

Multiply a file by the binary mask to create filtered output (e.g. calc_out). In this expression, we multiply the original epan image (study1_norm+orig) by our mask (outfimfunc+orig) and generate calc_out+orig. calc_out+orig is a sea of 0's except that the raw data from study1_norm+orig is preserved wherever outfimfunc+orig =1 (e.g., a square in the upper left corner of every timepoint in slice 5). Note that afni compensates for the fact that these two files are different sizes (outfimfico is 3d, but study1_norm is 3d+time). calc_out is 3d+time.

Data:

• Copy the talairached files into your working directory now

Procedure:

• Open afni and select "Talairach View" from the top right of the main panel.
• Choose Define Datamode->Plugins->Draw Dataset. The lower third of the Draw Dataset Panel is no longer greyed out because now tlrc images are available.
• Choose "Copy Dataset" Zero Func AsIs. (The dataset you will be drawing on will be a functional overlay named COPY_brain3d, a fim file)
• Select the brain3d image in +tlrc. Make sure that COPY_brain3d is displayed as the image you are working on.
• Now click on the axial, coronal and sagittal buttons on the main panel.
  • Make sure you are displaying brain3d in the images
• Right-click one of the axial, coronal or sagital images, and from the popup, select "Atlas Colors". This will take a little while to load.
• Select some distinctive colors for structures (e.g., red for hippocampus, green for amygdala).
• Name the file to save (e.g., test) and load it to make sure it will apply. Click "Done".
• From the 'Draw Dataset" Panel, choose TT atlas regions to load (e.g., Since we have designated distinctive atlas colors for them, choose "Hippocampus" and then "Load:OverWrite". Do the same for the amygdala.
• You will want to "save" the image. Click "Done" to close the Draw Dataset Panel.
• Choose "Define Function" (but NOT "See Function") and then on the lower right of the define function panel, choose "See TT atlas Regions".
• If you selected "See functions" then all of your color choices in the atlas are meaningless...colors will be selected from the intensity colorbar (so you'll probably see all your selections as yellow....turn "See Functions" off.
• It does not matter which functional file you have selected, because you are not viewing functions. If you have trouble, try clicking "See TT Atlas Regions" on and off to refresh the image. Make sure your color scheme is loaded (right click the image and choose atlas colors again to make sure your color chaoices are displayed). If this doesn't work, try restarting Afni. Be patient with yourself as this can be a little tricky, but quite worthwhile.
• Unless you reset the default "drawing value", it will be "1". This means that no matter how many anatomical structures you add to your picture, and no matter what pretty colors you choose for them, you are still creating a 1/0 binary mask in which filled anatomical structures have the value "1" and everything else is "0". You can use this binary mask with 3dcalc, just like you would use any other binary mask.

If you would like to view the functional image as a mask over a high resolution image, proceed as follows:

First, make sure anatomical images are aligned with the spirals (functionals). By default the 2Ds (T1's) are aligned with the spiral images. The 3D needs to be centered and aligned. See instructions for aligning the 3Ds. Run 3dAutomask on the preprocessed spiral image:

>3dAutomask study1_norm+orig

The result will be a fim image (typically 64x64xnas). Afni will name it automask* by default and treat it as a functional image.
>afni
Display the high resolution anatomical image, choose "See Function" and select your automask:

You could, of course, also multiply your functional activation image by the automask to remove all data outside the brain.

In SPM, imcalc is the analog of 3dcalc in afni. imcalc allows you to perform various algebraic functions on images. For example, imcalc will allow you to perform Matlab dot multiplication on two images with a command like this to imcalc:

i1 .* i2

(where i1 is the first image you select in the spm interface and i2 is the second image).

Matlab dot multiplication is exactly what is described above in the section on multiplication.

Marsbar, a freely downloadable and well documented toolbox for spm comes with a library of anatomical masks in mni space. These can be converted to img files and then multiplied with a subject's img file to get a mask of that subject's values for the particular area.

(e.g., convert the mni_left hippocampus file to an img file and dot multiply that image and your subject's functional image together to create an image of your subject's intensity values in the hippocampus only.)

See also: http://www.mrc-cbu.cam.ac.uk/Imaging/Common/epimasking.shtml