Assessing the accuracy of detecting mouse brain structure changes from MRI using simulated deformations

Introduction The use of image registration techniques to investigate shape differences in mouse brain MRIs have become a significant area of interest [1-3]. The ability of these techniques to bring brains into alignment have been well documented [4,5], however, it is unknown how accurately structural changes between groups can be detected or whether this sensitivity varies with brain geography or structure shape. Here we present a novel method to simulate deformation fields with known structural tissue shrinkage or growth and subsequently attempt to recover the induced changes in 21 structures of the mouse brain. Methods A simulated deformation field inspired by [6] was created as follows: using a 3D atlas of the mouse brain with 62 structures identified [7], a target Jacobian determinant map was created, in which any voxel of one structure received a reduced determinant (less than 1, i.e., tissue shrinkage), while the remaining part of the brain was assigned a determinant of 1 (no change). Concurrently, a tolerance map was created to indicate areas outside of the brain which were permitted to deform (i.e., grow/shrink) to accommodate the induced changes in the brain. A deformation field with zero vectors was initialized and iteratively adjusted by updating the vectors of each voxel’s six nearest neighbours until the resulting vector field yielded the same determinant map as was inputted, using the tolerance map to park volume changes in areas outside the brain. To test the ability of a registration algorithm to recover induced structural changes, 21 of the relatively larger structures of the mouse brain were chosen and deformation fields simulated which featured a structural tissue loss of 0.5 to 10 percent. A set of 20 identical wild type fixed brain MRI scans were then selected and the simulated deformation field applied to half of them. An iterative registration procedure, previously applied to multiple phenotyping studies [3,8,9], was then applied to these 20 brains and the resulting Jacobian determinants of the deformation fields analyzed for structural group differences. Multiple comparisons were accounted for using the False Discovery Rate (FDR) [10].