To find the brain, BSE applies a series of mathematical morphological operators to the edge map found in the edge detection step. The first of these is an erosion with a 3D cross (diamond) element, which will expand the edges (black voxels) into each voxel that shares a face with an edge voxel. The result is that small connections less than 2 voxels wide will be broken. For images with resolution of 1mm3, this corresponds to a 2mm structure – this is sufficient to erode most cranial nerves and to eliminate most noise-related connections between the brain and scalp portions of the MRI. With finer resolution images, it may be necessary to increase the erosion size to 2 in order to separate the brain.
After this erosion, BSE segments the eroded edge map into several connected regions that are bounded by the edges. BSE then analyzes these regions to select a candidate brain region based on (1) size of the connected region (2) the average intensity of the connected region and (3) the position of the centroid of the connected region.
- Erosion size (typical values are 1 and 2)
- Determines how far to extend edge (black) voxels into edges that share a face with them. The result is to break up/remove small connections that are less than twice the erosion size, which in turn helps remove cranial nerves and detected connections between the brain and scalp due to noise in the image. In most instances this can be left at 1, but high-resolution images might necessitate an erosion size of 2.