|M.Sc Student||Guy Nir|
|Subject||Temporal Registration of Partial Data Using Particle Filter|
|Department||Department of Electrical Engineering||Supervisor||Mr. Tannenbaum Allen Robert|
|Full Thesis text|
Image registration is the process of establishing a common geometric reference frame between two or more image data sets possibly acquired at different times, from different perspectives or by different imaging techniques. Registration is a problem of key importance in most image processing applications, including medical imaging.
In this work, we propose a particle filtering framework for rigid and non-rigid temporal registration of a model image to a time-series of partially observed images. The method tracks the pose and shape dynamics of an underlying observed object by employing a shape-based segmentation technique, and finds the corresponding elastic map to the model image using the Monge-Kantorovich theory of optimal mass transport. The method is robust with respect to image noise and clutter, variations of illumination and different image modalities.
A novel method for intensity-based non-rigid registration of partial data is introduced as well. An optimal region in the model image, corresponding to that of the partial data in the observed image, is found directly from the intensity values by minimizing the "Earth Mover's Distance", defined by the L2 Kantorovich-Wasserstein metric, among the corresponding regions.
Finally, an applicable algorithm is derived by employing the proposed framework to registration of consecutive axial 2D slices of a prostate magnetic resonance imaging (MRI) scan, acquired intraoperative over time, into a 3D model of that prostate, segmented from preoperative images. Such temporal slice-to-volume registration algorithm may be implemented to enable an existing transrectal prostate biopsy device to perform multi-parametric MRI guided prostate biopsy in closed-bore high field MRI magnets. We test the algorithm on various scenarios.