Face Anonymizer (BET)

Jump to: navigation, search

This Face Anonymizer (BET) algorithm extracts an approximated face from a T1-weighted MRI, eliminating facial areas outside of a certain buffer from the brain. The algorithm employs several other algorithms developed by the MIPAV project.


The procedure used by this algorithm to return an image with an extracted face consists of five major steps:

  1. Extract the brain.
  2. Buffer the brain.
  3. Define the face.
  4. Extract the face.
  5. Smooth the face.

Step 1, Extract the brain

The brain extraction portion of the algorithm employs the Extract Brain Surface (BET) tool. This tool extracts the surface of the brain from a T1-weighted MRI by expanding an initial ellipsoid. The brain extraction tool is derived from a similar tool developed at the Oxford Center for Functional Magnetic Resonance Imaging of the Brain.

Parameters that directly relate to the mechanics of the brain extraction tool can be changed using the dialog box for the face anonymizer. For more information, refer to the dialog help section.

For more information concerning the mechanics of this portion of the algorithm, see references.

Step 2: Buffer the brain

The brain extracted from the brain extraction tool is buffered by a certain number of millimeters that can be set using this algorithm's dialog menu. This portion of the algorithm employs the morphological dilate algorithm to extend the brain mask derived from brain extraction tool in three dimensions. Refer to Extract Brain: Extract Brain Surface (BSE) for more information.

Step 3: Define the face

The face is searched for in an area defined by the MRI given orientation. This orientation is derived from the file information of the MRI and can be changed using this algorithm's dialog options. Within this region an initial shape is chosen that will expand in an effort to extract the face.

Step 4: Extract the face

Face extraction is achieved by appending spheres of random radius onto the initial shape. This size of the spheres decays exponentially as the area of the two octants is filled. Eventually the process converges to a shape that is defined as an approximate face and does not lie within the buffered brain area. For more information on the method of sphere approximation, see references.

Step 5: Smooth the face

The extracted face is smoothed to approximate a quarter-cylinder within the two given octants. The final extracted face is guaranteed to not lie within the extracted brain.

Image types

You can apply this algorithm only to 3D T1-weighted MRI images. The image file must represent the whole brain (from the left to the right side and from the top to the bottom of the brain). The algorithm would not work if the image file contains only a few slices.


The resulting image is of the same data type as the original image.


Within the MIPAV project:

The brain extraction tool, which extracts the surface of the brain from a T1-weighted MRI and the dilate algorithm, which buffers the extracted brain by a set number of millimeters, are described here: Extract Brain: Extract Brain Surface (BSE).

Outside of MIPAV:

Smith, Stephen. BET: Brain Extraction Tool. FMRIB Technical Report TR00SMS2, Oxford Center for Functional Magnetic Resonance Imaging of the Brain, Department of Clinical Neurology, Oxford University, Oxford, England. Smith's paper is also available here: http://portal.acm.org/citation.cfm?id=246

McIlroy, M. D. Best approximate circles on integer grids. ACM Transactions on Graphics (TOG), Volume 2 Issue 4. An approximation for spheres also available at http://portal.acm.org/citation.cfm?id=246.

Applying the Face Anonymizer Algorithm

To use this algorithm, do the following:

  1. Select Algorithms > Brain Tools > Face de-Identification. The Face Anonymizer dialog box appears.
  2. Complete the information in the dialog box. Check that the MRI orientation information is correct.
  3. Click OK.
  4. The algorithm begins to run, and a progress bar appears with the status. When the algorithm finishes running, the progress bar disappears, and the original image without the extracted face appears.

Figure 1. An example of the Face Anonymizer (BET) image processing. The image has been processed using default parameters.



Recommendation: To find an optimal set of parameter values, run this algorithm repeatedly on a representative volume of the 3D T1-weighted MRI images using different parameter values. Keep the records of parameter values you use.

Figure 2. Face Anonymizer dialog box options
Which way is the patient's face pointing?
Specifies the orientation of the image. MIPAV extrapolates this value from the image file information. Confirm that the orientation given is correct:
Right - sagittal image with x-axis oriented posterior to anterior;
Left - sagittal image with X-axis oriented anterior to posterior;
Down - axial image with Y-axis oriented posterior to anterior;
Up - axial image with y-axis oriented anterior to posterior;
Into the screen - coronal image with Z-axis oriented posterior to anterior;
Out of the screen - coronal image with Z-axis oriented anterior to posterior.
Face Removal Options
Brain buffer
Guarantees that the extracted brain will be avoided by the specified number of millimeters. Refer to Step 2: Buffer the brain for more information.
Brain Extraction Options
Initial shape
Uses a sphere shape instead of an ellipse for estimating the initial boundary.
Specifies whether a sphere should be used for initial brain extraction. For more information, see the sphere approximation option for the brain extraction tool.
Image influence ratio
Specifies the depth to calculate in approximating brain maximum and minimum intensities. For more information, refer to "Extract Brain: Extract Brain Surface (BSE)" to learn more about the image influence option for the brain extraction tool. Default value is 0.1. Allowed values range from 0.01 to 0.5.
Mesh stiffness
Controls the stiffness of the brain approximation mesh. For more information about the stiffness option, refer to Chapter "Extract Brain: Extract Brain Surface (BSE)"of the MIPAV manual.
Default value is 0.15. Allowed values range from 0.01 to 0.5.
Confirms the dialog settings and runs the algorithm.
Disregards any changes that you made in this dialog box and closes this dialog box.
Displays online help for this dialog box.