Purpose: To propose and assess methods that permit the accurate and robust registration of serially acquired 3D MR scans and to demonstrate their application in neurodegenerative disorders.
Method: 3D T1-weighted brain images were obtained on two or more occasions from 10 normal subjects and 8 patients with neurodegenerative disorders (scan intervals 3-21 months). Variants of an automated registration procedure were compared according to the goodness and robustness of match in normal subjects. Consistency was measured using scan triplets. Change was visualized by "difference overlay images."
Results: A multiresolution method was necessary for robust registration. Significant improvements in the registering of control scans were produced with voxel size correction (p < 0.05), matching only brain voxels (p < 0.05), and sinc interpolation (p < 0.05). Fast sinc resampling was 20 times faster than an equivalent previous method. Subvoxel accuracy was demonstrated. Difference overlay images showed little cerebral change in normal subjects; in Alzheimer disease patients, characteristic patterns of brain atrophy were observed even with scan intervals as short as 3 months.
Conclusion: This methodology permits subvoxel comparison of routinely acquired serial 3D MR brain scans. It is a sensitive method for tracking patterns and rates of neuroanatomical change.