Group-wise motion correction of brain perfusion images

This work presents a group-wise motion correction method to align a series of brain perfusion scans simultaneously. The cost function used in this group-wise registration technique is derived from the total quadratic variation of image intensity. Optimization of the total quadratic variation with respect to rigid registration parameters is a least squares problem and can be efficiently solved by the Levenberg-Marquardt method. The group-wise registration error was quantitatively measured by comparing the registration results with ground truths that were obtained by applying artificial motions to a series of 50 brain perfusion scans. The average registration error is 0.869 ± 0.579 mm, and the maximum error is 3.550 mm. The capture range and robustness was also tested by 11 MR brain perfusion scans with large patient motion. The results demonstrated a high level of accuracy and robustness achieved by the group-wise registration.