Key points are not available for this paper at this time.
Brain development, aging, and neurological and psychiatric disorders are often associated with structural changes of the brain. We describe here a fast and robust method for accurately quantifying cerebral structural changes, based on serial MRI scans. Longitudinal images are registered by first affine aligning and intensity-normalizing them with respect to each other. Given the target image with N voxels, the problem is to find the 3N-dimensional displacement field that locates corresponding intensities in the other, deforming, image. The displacement field is constrained to be smooth—local relative variation penalized—and stretched or compressed at each point by the intensity difference between the images. It is found by minimizing a merit function for the whole system that describes locally the linear elastic environment. To facilitate minimization, the images are heavily smoothed so as to make them similar, i.e., so that the starting point is in the basin of the global minimum of the merit function. Registration (minimization) is then repeated with reduced smoothing, starting from the updated displacement field. The whole procedure is iterated for more precise registration. Small ROIs can be zoomed for structures separated by only a voxel. The net displacement field allows volume changes to be calculated, and can be combined with automatic segmentation for ROI-specific values. The method has been applied to serial scans, including normal controls (NC) and subjects with Alzheimer's disease (AD), to determine volume changes of several ROIs. The most significant 12-month effect size, Cohen's-d of -2.2 for AD vs. NC, was observed in the lateral temporal cortex. For the left hippocampus, the Cohen's-d effect size was -1.73.
Holland et al. (Tue,) studied this question.