Software - Brain Boundary Shift Integral

The remarks on this page and our implementation of the BBSI algorithm are based on this paper by Freeborough and Fox.

The Brain Boundary Shift Integral is designed to compute volume changes due to the shifting of brain-csf boundaries over time. The basic idea is that if a voxel has the tissue intensity of white or gray matter at an earlier time and in a later scan has the intensity of CSF, then the tissue boundary has shifted at that location. These voxel intensity changes can be tallied and used to compute an estimated change in the overall volume of brain tissue between images at time 1 and time 2.

While the idea is simple, the computations can be delicate. The mean brain intensities of each of the images must be normalized. A mask is created to straddle tissue boundaries but preclude difference computations at locations away from boundaries. A window is imposed such that a voxel intensity difference between time 1 and time 2, greater than the window width, is clamped to that width. Similarly, voxel intensity changes smaller than a thresholded fraction of the window width are ignored as noise. The delicacy in this algorithm comes from proper settings of the window width, boundary mask and noise levels.

The figure illustrates the results of a BBSI computation. Voxels colored yellow have a higher intensity at time 1 than at time 2, thus indicating a boundary recession at the later time. Some orange-colored voxels indicate a higher value at time 2, either due to noise or a boundary shift in the opposite direction. The total volume change due to boundary shifts was 7.2 cc.

Figure Illustration of BBSI computations. Yellow voxels indicate boundary shifts due to expansion of CSF from time 1 to time 2. Orange indicates the reverse, but could also be due to noise. The BBSI shown here is from images taken within weeks of one another, so all effects are small and subject to distortion by noise.

Up (Software)