Registration of ultrasonography sequences based on temporal regions

2D Ultrasonography images with parallel contrast enhanced sequences for analysis constitute a rapid and inexpensive imaging technique with high temporal resolution to assess perfusion of tissue. However, motion from various influences corrupts the inter-pixel correspondences between different time frames and therefore hampers computer-assisted analysis of perfusion parameters. We present a user-supported method applying a temporal similarity matrix to remove frames with uncorrectable out-of-plane motion. For the remaining regions of frames, motion influence can be compensated for by image registration. Subsequently B-Spline based registration is applied using the temporal regions with automatic determination of a suitable reference frame image. Evaluation with ground truth data of six datasets comparing a medical expert frame analysis to the proposed technique yields 85.1% sensitivity and 81.7% specificity in average. On average 6% of the frames have been erroneously included in temporal regions, although they contain out-of-plane motion.