Automatic de-Skulling

From MIPAV
Revision as of 18:17, 16 January 2014 by Olga Vovk (Talk | contribs)

Jump to: navigation, search

Automatic de-Skulling provides tools that help anonymizing the 3D head image, so that the person cannot be recognized when displayed in a 2D-slice or 3D-volume view. This can be done as a progression from basic face-anonymizing, to de-skulling, to complete brain segmentation and extraction. Fully automatic brain segmentation and extraction is an area of on-going research.

Introduction

Full-scale de-skulling requires removal of all skin, muscle, skull, eyes, exterior blood vessels, and nerve tissues. Brain segmentation goes further to define the exact components (white matter, gray matter, cerebellum, brain stem) to include in the final image. Although brain segmentation is not necessary to de-identify a face, several components of brain segmentation algorithms can be used as strategies for removing the identifying features of a face and a skull.

Background

In this algorithm we decided to err on the side of leaving too much data rather than removing data that was part of the brain. One of the key steps in the algorithm is segmenting the white-matter of the brain,using a technique described in the Skull-Striping algorithm (Sadananthan, 2010). Given a good white-matter segmentation, the algorithm can reliability find a good skull segmentation without removing portions of the surface of the brain. A modified version of the Brain Extraction Tool (AlgorithmBrainExtractor) in MIPAV is then applied to further refine the brain surface.

The algorithm was tested on nearly two dozen images and on several examples of the different MRI modalities. The algorithm produced good segmentation for all the images except CT images. Refer to the Results section to view anonymization examples.

For the CT images the algorithm provides the option of using an atlas or reference image to generate the segmentation. The atlas image is first registered to the target image, then the algorithm is run on the atlas image and the result is applied to the target image. Using a T1-weighted MRI image as the atlas image worked well for the all CT images tested.

The goal for this algorithm is to produce fully anonymous results for

  • a variety of image types and modalities,
  • all age ranges from children to elderly,
  • both healthy and non-healthy patient images.

Algorithm Implementation

The algorithm performs the following steps:

  1. It calculates the input image histogram.
  2. It finds all pixel values that represent 2% and 98% of the histogram values from Step 1, and then clamps the input image data to that range.
  3. It sets i) the cerebrospinal fluid (CSF) threshold minimum value to 30 % of the clamped image range and ii) the CSF threshold maximum value to 70% of the clamped image range from Step 2.
  4. It calculates the brain center of gravity (COG) as the weighted sum of all voxels between the CSF minimum and maximum range using the voxel position scaled by intensity value.
  5. It estimates the white matter intensity threshold by i) segmenting the image into 5x5x5 voxel cubes and ii) finding the cube with the highest mean intensity and lowest variance. The mean intensity of the selected cube is used to set the white matter threshold value.
  6. It performs a white-matter segmentation of the brain by using a flood-fill algorithm with the white matter threshold seed point from Step 5 as the starting point. The flood-fill algorithm is limited by the CSF threshold values as well as a radius centered around the initial seed point.
  7. It takes the convex-hull of the white matter segmentation and passes this mesh into a modified version of the AlgorithmBrainExtractor.
  8. The output of the AlgorithmBrainExtractor is a segmented mesh representing the surface of the brain. To err on the side of not removing too much data, the convex-hull of the brain surface is generated.
  9. The final mask representing the data outside the brain is generated using all voxels outside the convex-hull of the brain surface.
  10. If the user chooses to segment only the face region instead of the entire skull region the final step produces the face segmentation from the convex-hull of the brain surface.

Steps 1-4

Are derived from the image pre-processing steps described in the Extract Brain Surface tool (BET and BET2) (Smith, 2000). The preprocessed steps generate the center of gravity (COG) and cerebrospinal fluid (CSF) thresholds, which are used to find the initial white-matter seed point. Potential white matter seed points are limited to a radius around the COG. The minimum variance is selected from potential seed points with a maximum mean-intensity greater than the CSF threshold. These steps prevent the algorithm from finding a seed point in the background, where the variance may be zero but where the intensity is lower than the CSF threshold.

Steps 5-6

Are based on the Skull-Striping algorithm (Sadananthan, 2010). Once the white-matter seed point is found it is used to initialize a flood-fill algorithm that segments the white-matter of the brain. The seed point is typically found in one hemisphere of the brain. The first step in initializing the flood-fill algorithm is to generate a mirror-image seed-point located in the opposite hemisphere. This enables the flood-fill algorithm to work equally well on both hemispheres. The flood-fill algorithm is limited to an ellipsoid shape around the COG, where the long axis of the ellipsoid is anterior-to-posterior and the short axes of the ellipsoid are superior-to-inferior and left-to-right.

Step 7

Once the white-matter has been segmented, a convex-hull of the segmentation is generated in Step 7. This mesh is passed to a modified version of the AlgorithmBrainExtractor in MIPAV. The convex-hull of the white-matter often provides a good enough segmentation of the skull, but in a few cases the convex-hull of the white-matter does not include the entire surface of the brain and processing the mesh with the AlgorithmBrainExtractor corrects the mesh.

Steps 8-9

Were added to provide a buffer around the brain segmentation. This enables the algorithm to err on the side of removing too little data rather than removing small portions of the brain surface.

Step 10

If the user chooses to segment only the face region, the face mask is generated from the convex-hull of the brain. To generate the face mask, the center of gravity (COG) is used to generate two points on the surface of the mesh. A line parallel to the x-axis centered at the COG intersects the convex-hull, and the intersection point farthest from the COG is selected as the first point on the surface. A second line parallel to the y-axis and centered at the COG intersects the convex-hull and the intersection point farthest from the COG is selected as the second point on the surface.

The two points on the hull define a clipping plane that intersects and may cut through the convex-hull. The clipping plane is then moved through the convex-hull to find the point where it is tangent to the surface and does not cut through the surface. The plane is truncated where it intersects the two lines from the COG parallel to the x and y-axes. Truncating the plane prevents the spinal column from being clipped.

Results

The success of the algorithm depends on the white-matter segmentation that initializes the modified BET algorithm. The white-matter segmentation proved to be very robust across the different image types, as shown in the images below.

Only CT data proved difficult for the algorithm to accurately segment the white-matter. To better de-skull the CT data the user must load an atlas image as part of the algorithm. The atlas image option is described in the Applying the algorithm section.

Image types

You can apply this algorithm to 3D MRI and CT images.

Input Image Challenges

A significant challenge in creating the algorithm is the variety of input images.

  • Computed Tomography (CT) [1].
  • Magnetic Resonance Images (MRI) [2]including but not limited to:
    • Diffusion-weighted (DWI) [3],
    • Diffusion-tensor (DTI) [4],
    • High-angular-resolution (HARDI) [5],
    • Fluid attenuated inversion recovery MRI (FLAIR)[6],
    • Functional MRI (fMRI) [7],
    • Proton density MRI (PD) MRI,
    • T1-weighted MRI [8],
    • T2-weighted MRI [9].

Much of the reference work in brain segmentation is done on T1 or T2 images. Examples of different image types are listed and shown in the Results section.

Results for face anonymizing

T2 images

T2W VISTA(3DT2) 601.jpg The original T2 image
T2W VISTA(3DT2) 601BlurFace.jpg T2W VISTA(3DT2) 601RemoveFace.jpg T2W VISTA(3DT2) 601RemoveSkull.jpg
The image with the face blurred The image with the face removed The image with the skull removed
T2W VISTA(3DT2) 601WhiteMatter.jpg T2W VISTA(3DT2) 601FaceSegmentation.jpg T2W VISTA(3DT2) 601SkullSegmentation.jpg
White matter segmentation Face segmentation Skull segmentation

Applying the algorithm

To use this algorithm,

Do the following:

  1. Open a brain image which you would like to anonymize. In [[Opening and loading image files|the Open Image dialog box], check the Open as Multifile option.
  2. Select Algorithms > Brain tools > Anonymize. The Anonymize face dialog box opens.
  3. Choose one of the following face anonymizing options: Blur face, Remove face, Show face segmentation results, Blur skull, Remove skull, Show skull segmentation results.
  4. Complete the rest of the information in the dialog box. See Dialog box options.
  5. Click OK. The algorithm begins to run, and the progress bar appears with the status. When the algorithm finishes running, the progress bar disappears, and the result appears in a new image window.

To use the algorithm with an atlas image,

Using the atlas image is recommended for CT images. Using a T1-weighted MRI images as the atlas images worked well for the all CT images tested.

To run the algorithm, do the following:

  1. Open a brain image which you would like to anonymize. In [[Opening and loading image files|the Open Image dialog box], check the Open as Multifile option.
  2. Select Algorithms > Brain tools > Anonymize. The Anonymize face dialog box opens.
  3. Choose one of the following face anonymizing options: Blur face, Remove face, Show face segmentation results, Blur skull, Remove skull, Show skull segmentation results.
  4. Choose the atlas image: in the Optional Atlas Image box, use the Browse button to select the atlas image from your disk. Check the Open as Multifile option in the Open Image dialog box.
  5. Complete the rest of the information in the dialog box. See Dialog box options.
  6. Click OK.

The algorithm begins to run. The atlas image appears in a new image window.

When the atlas image provided, the algorithm does the following:

  1. It registers the atlas image to the original (target) image,
  2. It runs the selected anonymization option on the atlas image,
  3. It applies the result from Step 3 to the target image.

As the algorithm starts the anonymization routine, the progress bar appears with the status. When the algorithm finishes running, the progress bar disappears, and the result appears in a new image window.


Running the Automatic De-Skulling algorithm with an atlas image

Progress bar note

AutomaticDeSkullingProgressBar.jpg

The progress bar may slow down as it reaches 60%, just be patient, the algorithm is still working.

Atlas image note

If one image is acquired from a patient/subject, and the other image is taken from an image information database, we call it atlas registration. In the literature, registration of a patient image to an image of a normal subject is also termed atlas registration.

Dialog box options

Anonymize face dialog box
AnonymizeFaceDialogBox.jpg
Blur face The algorithm first computes and then blurs the face mask voxels. The blurred face volume is copied into the original image and displayed in a new image window.
Remove face The algorithm i) computes the face mask voxels and ii) removes them from the original image. The result image is displayed in a new image window.
Show face segmentation results The algorithm computes the face mask voxels. The segmentation results including the brain white matter segmentation and face segmentation are displayed in a new image windows.
Blur skull The algorithm first computes and then blurs the skull mask voxels. The blurred skull volume is copied into the original image and displayed in a new image window.
Remove skull The algorithm i) computes the face mask voxels and ii) removes them from the original image. The result image is displayed in a new image window.
Show skull segmentation results The algorithm computes the skull mask voxels. The segmentation results including the brain white matter segmentation and skull segmentation are displayed in a new image windows.
Face removal options
Extracted brain is avoided by a buffer of this many mm Enter the number (in mm) to ensure that the extracted brain will be avoided by the specified number of millimeters. By default it is set to 20.
Optional Atlas Image To speed up the process and ensure accuracy, the user can use an atlas image that generates a good segmentation.
OK Runs the algorithm.
Cancel Disregards any changes that you made in the dialog box and closes this dialog box.
Help Displays online help for this dialog box.


References

  • Pechaud M., Jenkinson M., and Smith Stephen M., FMRIB Technical Report TR06MP1 : MRI-Based Estimation of Brain, Skull and Scalp Surfaces, 2006, PDF
  • Sadananthan et al., Skull stripping using graph cuts, NeuroImage 49, 225-239, 2010, PDF
  • Stephen M. Smith, FMRIB Technical Report TR00SMS2: BET: Brain Extraction Tool, Oxford Centre for Functional Magnetic Resonance Imaging of the Brain, 2000, PDF