Dynamic magnetic particle imaging: accurate reconstructions by simultaneous motion estimation

Abstract

Magnetic Particle Imaging (MPI) has a particularly high spatial and temporal resolution. This temporal resolution makes it a very appropriate candidate for imaging dynamic tracer material and the dynamics itself inside the body. Potential applications include instrument tracking during interventions, but also blood flow imaging, where the dynamics do not only exist but are of high interest as they can be used directly for diagnostic purposes.

However, the image reconstruction task poses a severely ill-posed inverse problem even for static tracer concentrations and we face an even more challenging problem in case of dynamic concentrations. More particularly, we have to accept inaccuracies in our forward modeling in order to obtain fast reconstruction algorithms. Moreover, we expect severe motion artifacts in the reconstructed images due to the non-instantaneous measurements in MPI for two reasons: the motion might exceed one voxel per time frame and the possibility of averaging over frames in order to increase the data SNR is very limited.

In this poster, we propose to solve the dynamic image reconstruction task jointly with motion estimation in between the time frames, as both processes endorse each other and motion estimates are of interest in many dynamic applications [1]. We use different motion models depending on the specific application and start from a fairly general variational problem formulation. The problem is solved by primal-dual splitting using stochastic algorithms, multi-scale approaches and image warping. We present convincing numerical results on measured data.

[1] M. Burger, H. Dirks, and C.-B. Schönlieb. A variational model for joint motion estimation and image reconstruction. SIAM Journal on Imaging Sciences, 11(1):94–128, 2018.