CT brain extraction

I want to remove the skull from CT images. I have been through SkullStripping.cxx code. I do not have a brain mask. Instead, I have skull mask. Is there any other resource that I could look into?

Getting the mask for the outside of the skull should be easy - simple thresholding and keeping the largest component should do. Then combine skull and air, invert, and you have your brain mask.

1 Like

Thank youā€¦What do you mean by ā€˜airā€™ here?

Air outside of the head. Well, also the bed on which the subject is lying.

Skull stripping of CT images is a much easier problem than MR images, because the bone can be segmented using a simple global threshold operation. This allows using methods that do not require an ā€œatlasā€ image, making the segmentation method more robust and the results more accurate. The main challenge that remain is to prevent leaking out through large holes on the skull, such as foramen magnum and through smaller holes that appear on the skull (other cranial foramina are quite small, but larger holes may appear on CT at thin bone segments, such as orbital wall, due to partial volume effect).

We have recently compared purely bone-segmentation-based CT skull stripping methods (which are applicable to even dry-bone CTs, where there is no contrast between the air and the brain). You can find the results here.

We then continued to explore and simplify the fully automatic approach and come up with a robust and fully automatic cavity segmentation method based on inverted shrink wrapping:

The algorithm is available in 3D Slicerā€™s Segment Editor after installing SurfaceWrapSolidify extension. It can be run fully automatically from Python script within Slicer or you can take the source code and use it in your own software.

1 Like

I had published on something that used FSL in the past: https://www.sciencedirect.com/science/article/pii/S1053811915002700?casa_token=6ha0nal7gU0AAAAA:1EgaY3WeOCJZnTdoSBK1yQ-zTYvQKSucQSS-CBdOI7YcMAQgsaW9dF4tCLQQazEFDP8Gb486UA

The procedure is simply, thresholding based on Hounsfield Units, some small amounts of smoothing, and then running the Brain Extraction Tool (BET). There are some specific issues with certain cases, but it works 95% of the time.