Application of 3D‐MR image registration to monitor diseases around the knee joint

Purpose To estimate the accuracy and consistency of a method using a voxel‐based MR image registration algorithm for precise monitoring of knee joint diseases. Materials and Methods Rigid body transformation was calculated using a normalized cross‐correlation (NCC) algorithm involving simple manual segmentation of the bone region based on its anatomical features. The accuracy of registration was evaluated using four phantoms, followed by a consistency test using MR data from the 11 patients with knee joint disease. Results The registration accuracy in the phantom experiment was 0.49 ± 0.19 mm (SD) for the femur and 0.56 ± 0.21 mm (SD) for the tibia. The consistency value in the experiment using clinical data was 0.69 ± 0.25 mm (SD) for the femur and 0.77 ± 0.37 mm (SD) for the tibia. These values were all smaller than a voxel (1.25 × 1.25 × 1.5 mm). Conclusion The present method based on an NCC algorithm can be used to register serial MR images of the knee joint with error on the order of a subvoxel. This method would be useful for precisely assessing therapeutic response and monitoring knee joint diseases; normalized cross‐correlation; accuracy J. Magn. Reson. Imaging 2005. © 2005 Wiley‐Liss, Inc.