Image Reconstruction And Registration For Liver 4D Dynamic Contrast-enhanced Magnetic Resonance Imaging

Hepatocellular carcinoma is one of the most heterogeneous malignancies with the highest mortality rate.In the process of cirrhosis nodules gradually formed the hepatocellular carcinoma,the increased newborn abnormal arterial blood supply is an important criterion for clinical diagnosis.Liver four-dimensional dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)can not only provide the morphological information of lesions,but also can reflect the changes of blood microcirculation of pathological tissues,and thus providing more abundant and accurate image information for the early diagnosis and curative effect evaluation of hepatocellular carcinoma.The clinical application of liver 4D DCE-MRI is faced with two major technical challenges:the first is rapid imaging,which needs to reconstruct the high spatial and temporal resolution of the whole liver 4D image based on highly dersampled k-space data;The second is respiratory motion correction,which needs to remove the artifacts caused by respiratory motion during the long scanning time effectively.This paper aims to reseach the key technologies of whole liver 4D DCE-MRI,and the innovative achievements include:(1)A dictionary learning based low-rank plus sparse decomposition reconstruction algorithm is proposed to reconstruct the high spatial and temporal resolution images based on the highly undersampled k-space data.The low rank sparse decomposition model is used to effectively mining the correlation between reconstructed image frames,and the dictionary learning is used to further enhance the adaptability of the algorithm,thus realizing the high sparse expression of the image signals.Experiments show that our algorithm can effectively remove the undersampling artifacts and improve the reconstructed images' details.(2)A dynamic MRI reconstruction algorithm with respiratory motion correction is proposed for free breathing scans.Firstly,the one-dimensional breath motion signal is directly estimated based on the k-space data,and the k-space data is divided into several motion states according to the signal.Then,the image sequence is reconstructed by low-rank sparse decomposition based on the k-space data of each state.Lastly,the sparse constraint is enfoced on the entire sequence to achieve the second reconstruction.Experiments show that the proposed algorithm can effectively remove motion artifacts and reconstruct the high spatial and temporal resolution images under free breathing.(3)Aiming at the problem of the respiratory motion registration of the chest and abdomen images,a discontinuity preserving non-rigid registration is proposed.The algorithm use a Markov random field based discrete optimization method.The interface between the viscera and the inner wall of the chest and abdomen is automatically segmented with the estimated motion field and the image signal,and then the smoothness constraint of the motion field across the interface is removed.Experiments show that this algorithm can achieve the automatic segmentation of the respiratory sliding interface and reduce the registration error effectively.