Session P7F.3

Half K-Space Reconstruction Using the Analytic Image Concept

J Yankam Njiwa*, YM Zhu

CREATIS-INSA
Lyon, France

Partial k-space acquisition is a subject of great interest in MRI. This type of acquisition usually consists of, in addition to the positive half of k -space, acquiring a number of negative frequencies beyond the k-space center, thus requiring an increase of acquisition time. Furthermore, the quality of the reconstructed image depends on the number of negative frequencies acquired. We propose a new approach for reconstructing magnetic resonance (MR) images from half k-space. It is based on the notion of analytic image and presents the particularity of using only exactly half of k-space acquired in an optimal asymmetrical schema.
The need of higher acceleration factor by just acquiring a part of the k-space leads to an amplification of some artefacts or noise during the reconstruction. The proposed method exploits redundancies in the k-space data and according to the asymmetrical acquisition schema, the benefits of the analytic image concept with the knowledge of only a part of the high spatial frequencies. It is a good trade-off between the acquisition and the quality of the reconstructed image both in the qualitatively and quantitatively meaning. The reconstruction process consists in splitting a given k-space into two parts, the positive and the negative spatial frequencies. The two parts are then considered as the spectral representations of two analytic images. The final image is then reconstructed by taking the magnitude of the sum of the two reconstructed analytic images.
The proposed MR reconstruction method was evaluated on both human and phantoms data. For the human MR data, the raw k-space data were acquired on a clinical 1.5T whole body Siemens Sonata system (Erlangen, Germany) and the phantom data were acquired on a Bruker 7T. The reconstruction quality was assessed both qualitatively (difference images) and using quantitative criteria such as normalized root mean square error (NMSE) and signal to noise ratio (SNR). The obtained results showed that, with the conventional POCS method, the reconstruction quality is characterized by the presence of some artefacts like some aliasing on the edges of the images, whereas with the proposed method, the reconstruction quality is fairly good and presents good SNR and smaller NMSE values. Moreover, in all cases, the proposed method exhibited a substantially better reconstruction quality than the POCS method. So, the proposed method based on analytic image representation allows achieving good MR images reconstruction quality using only one and exactly one half of k-space data without involving any phase correction, thus reducing the acquisition time by a factor of 2.

(Abstract Control Number: 243)