A locally adaptive registration technique for high precision registration of 3-D MRI data

This study demonstrates how the rigid body registration parameters for good registration of serially acquired 3-D magnetic resonance images vary systematically when the registration routine is presented with a series of cropped data sets that are systematically positioned throughout the entire volume. The results of the registration of these subcubes are compared with the results of a single registration of the complete volume for two consecutive 3-D scans of the brain of a normal volunteer, with one scan having optimized shim coil currents and the other having all second-order shim coil currents set to zero. The technique is sensitive and able to reveal subvoxel misregistrations.