Software - T1 Skull Strip

The skull stripping application works on T1 images to remove the skull. User intervention may be required usually just for minor cleanup afterwards. First the skull-strip estimates tissue boundaries by finding inflection points (zero points of the second derivative) in the image histogram. Then it uses a “watershed algorithm” to fill the compartments defined from these boundaries, resulting in a brain that is usually fairly clean of surrounding dura or attachments at the base of the brain.

For minor cleanups we use a version of the slice viewer that displays the stripped brain overlaid in color on the whole head image. A brush of adjustable size can then either erase small areas that jut from the brain or restore small chunks that have been mistakenly removed during the skullstrip.

It should be emphasized, however, that the product of the fully-automated skull-stripping phase is often good enough for most purposes. We and another lab have compared it with the commonly used Brain Extraction Tool. It appears that the two algorithms have complementary behavior, in that one may succeed where the other fails. When our skullstrip fails in a big way, this is usually traced to an image histogram that has poor separation of tissue intensities.

Figure Automated T1 image skull stripping. Left: unstripped image. Right: stripped image after automated stage and minor user cleanup.

Up (Preprocessing)

7588 unique visits since June 1, 2006