For sequential studies of patients with brain tumors, the authors have
designed an automated registration procedure for intra- and interexam
ination alignment of magnetic resonance images. This was evaluated wit
h artificially misregistered data and data from repeat studies of six
healthy volunteers and six brain tumor patients. In a subset of cases,
a manual procedure based on matching of neuroanatomic landmarks was a
lso applied for comparison. The results showed that the technique is r
obust and reproducible, giving an accuracy in the range of 1-2 mm, whi
ch corresponded to the spatial resolution of the images. Subject motio
n between imaging sequences within the same study was negligible, alth
ough adjustments (one to two section thicknesses) were required in the
z direction to correlate multisection and volume images and to allow
accurate image segmentation. For alignment between sequential voluntee
r and patient examinations, translations of up to 22 mm and rotations
in the x, y, and z axes of up to 9 degrees were required. This alignme
nt procedure may be valuable in numerous aspects of treatment planning
and patient follow-up.