Joint multi-patch reconstruction: fast and improved results by stochastic optimization

In order to measure larger volumes in magnetic particle imaging, it is necessary to divide the region of interest into several patches and measure those patches individually due to a limited size of the field of view. This procedure yields truncation artifacts at the patches boundaries during reconstruction. Applying a regularization which takes into account neighbourhood structures not only on one patch but across all patches can significantly reduce those artifacts. However, the current state-of-the-art reconstruction method using the Kaczmarz algorithm is limited to Tikhonov regularization. We thus propose to use the stochastic primal-dual hybrid gradient method to solve the multi-patch reconstruction task. Our experiments show that the quality of our reconstructions is significantly higher than those obtained by Tikhonov regularization and Kaczmarz method. Moreover, using our proposed method, a joint reconstruction considerably reduces the computational costs compared to multiple single-patch reconstructions. The algorithm proposed is thus competitive to the current state-of-the-art method not only regarding reconstruction quality but also concerning the computational effort.

DDC Class

004: Informatik

570: Biowissenschaften, Biologie

Projekt(e)

Modellierung, Simulation und Optimierung mit fluiddynamischen Anwendungen

Options

Joint multi-patch reconstruction: fast and improved results by stochastic optimization