Compressed Sensing Reconstruction Of High Spatio-temporal Resolution Abdominal Dynamic Magnetic Resonance Perfusion Images With Large Respiration Motion

Magnetic resonance imaging (MRI) has been widely used in clinical diagnoses and scientific research with prominent advantage, such as no radiation, multiplanar imaging, providing abundant pathological information and high contrast resolution in soft tissue. And dynamic MRI plays an important role in the field of cardiac cine, perfusion and functional MRI to capture the subtle changes in vivo. But due to scan time constraint, it is hard to reconstruct the dynamic image sequences with high temporal-spatial resolution and high signal-to-noise ratio. And respiration during the MRI scan will cause nonrigid motion of organ, which ruin k-space data consistency and bring motion artifacts. Multiple breath-holds is a common respiratory control protocol in dynamic MRI scan. But subjects usually have large respiratory motion after a long time breath-hold, which lead to a serious decline in the quality of image. In this paper, we study the dynamic MRI reconstruction technique based on the highly reduced sampling data for large respiratory motion. The main achievements of this paper include:(1) For undersampled dynamic MRI, a dictionary learning algorithm based on 3D spatiotemporal patches was presented. This method combined the double sparse model and high coherence in spatial and temporal domains to train a 3D dictionary. Compared with single level dictionary learning-DLMRI and the state of the art method k-t FOCUSS, our method performs better in capturing local structure and removing aliasing artifacts.(2) Markov random field (MRF)-based discrete optimization strategy had been proposed to overcome problems involved with B-spline free-form deformations (FFD) for registration. Experiment results demonstrate that the proposed method not only can cope well with the large and discontinuous sliding motion, performance indicators are also better than deedsMST and elastix registration algorithm.(3) With the dictionary learning and respiratory motion registration algorithm above, respiratory motion dynamic MRI reconstruction of under-sampled data for abdominal perfusion was implemented. We validated the improvements on multiphase spiral LAVA of human abdominal data. Compared with TRACER and PROUD reconstruction algorithm, our method can get better reconstruction quality of 3D MR images with high spatial and temporal resolution.This paper proposed a dynamic MR reconstruction system for abdominal perfusion, which can achieve sub-second frame rate with high quality 3D MR images from approximately 50 fold undersampling of k-space data under the influence of respiratory motion and contrast enhanced. Data collection is very simple in the system without respiratory gating and other techniques. In the reconstruction, the key algorithm modules were improved in many aspects. After sufficient experimental verification and compared with the existing advanced algorithms, the system can meet the application requirements.