Fully automatic nonrigid registration-based local motion estimation for motion-corrected iterative cardiac CT reconstruction

Methods: Using a gated four-dimensional region of interest image data set, a fully automatic elastic image registration is applied to recover a cardiac displacement field from a reference phase to a number of phases within the RR interval. Here, a stochastic optimizer and multiresolution approach are adopted to speed up the registration process. Subsequently, motion-compensated iterative reconstruction using the determined motion field is carried out. For the image representation volume-adapted spherical basis functions (blobs) are used to take the volume change caused by a divergent motion vector field into account. Results: The method is evaluated on phantom data and on four clinical data sets at a strong cardiac motion phase. Comparing the method to standard gated iterative reconstruction results shows that motion compensation strongly improves the image quality in these phases. A qualitative and quantitative accuracy study is presented for the estimated cardiac motion field. For the first time a blob-volume adaptation is applied on clinical data, and in the case of divergent motion it yields improved image quality. Conclusions: A fully automatic local cardiac motion compensated gated iterative method with volume-adapted blobs is proposed. The method leads to excellent motion-corrected images which outperform nonmotion corrected results in phases of strong cardiac motion. In clinical cases, a volume-dependent blob-footprint adaptation proves to be a good solution to take care of the change in the blob volume caused by a divergent motion field.