Motivated by previous work [16] and recent diffeomorphic image registration developments in which the characteristic velocity field is represented by spatiotemporal B-splines [3], we present a diffeomorphic B-spline-based image registration algorithm combining and extending these techniques. The advancements of the proposed framework over previous work include a preconditioned gradient descent algorithm and potential weighting of the metric gradient permitting, among other things, enforcement of stationary boundary conditions. In addition to theoretical and practical discussions of our contribution, we also describe its parallelized implementation as open source in the Insight Toolkit and perform an evaluation on publicly available brain data.
展开▼