Segmented Multistage Reconstruction of Magnetic Resonance Images

Abstract

Compressed sensing of magnetic resonance imaging refers to the reconstruction of magnetic resonance images from partially sampled k-space data. The k-space data reduces reconstruction processing time but at the cost of increasing artifacts - especially with the higher reduction factor of the raw data. This work proposes a segmented region-based reconstruction technique to reduce image artifacts with enhanced quality and high temporal resolution. The proposed method segments partially sampled k-space data in two segments according to their frequencies. Lower frequency components at the central region are selected and predicted using nuclear norm minimization. This part and the peripheral part of the k-space components at higher frequencies are merged. The recovery technique iterates to reconstruct more accurate images in terms of conventional compressed sensing techniques. The performance of the proposed method is evaluated and compared with compressed sensing, two-stage compressed sensing, and modified total variation technique. Better results in term of normalized mean square error NMSE, reconstruction time and structural similarity index measure SSIM show the effectiveness of the proposed method with a high reduction factor of data.

Keywords

• compressed sensing
• fourier transforms
• image reconstruction
• magnetic resonance imaging
• spatial resolution