Research On Method And Application Of Quantitative Susceptibility Mapping In MRI

Magnetic resonance imaging(MRI)that has the advantages of no radiation,multi-contrast and so on,is widely used in disease diagnosis and scientific research.Quantitative susceptibility mapping(QSM)developed in MRI is able to obtain the distribution of magnetic susceptibility in biological tissue,and has bright future of application in calcium content measurement.The typical characteristic of osteoporosis is calcium loss of bone tissue.In theory,the loss of diamagnetic calcium will lead to increased susceptibility of bone tissue.Therefore,QSM has great potential in osteoporosis assessment.In addition,the field-to-susceptibility dipole inversion that suffers from streaking artifacts in QSM reconstruction,remains an open question.In this paper,we did three main works around the research on spine QSM technology,the value of spine QSM in osteoporosis assessment,and the new method of field-to-susceptibility dipole inversion as follows:(1)Spine QSM method using ultrashort echo time and its application in osteoporosis assessment.QSM that is able to measure the magnetic susceptibility of biological tissue,which is closely related to bone mineral content,has the potential to provide a non-radiative indicator for osteoporosis assessment.In the implement of spine QSM,ultrashort echo time sequence was applied in data acquisition to obtain images with high signal noise ratio.In addition,avoiding the interference of fat signal in field map estimation,this work proposed to acquire in-phase echoes.After spine QSM using ultrashort echo time was successfully implemented,this work investigate the feasibility of spine QSM for osteoporosis assessment.The results showed that mean QSM in the osteoporotic cohort(82.0±39.9 ppb)was significantly higher than that in the osteopenic(30.8±47.0 ppb)and normal(-17.0±43.6 ppb)cohorts and QSM was negatively related to bone mineral density(r=-0.70).This work also investigated the scan-rescan reproducibility of spine QSM.The results showed that there were strong correlations(r=0.87?0.91)between two scans for QSM in L1?L4.(2)Spine QSM method using multi-echo gradient-echo with two repetition times.The spine QSM sequence using ultrashort echo time required about 10 minutes for data acquisition,and could not obtain proton-density fat fraction(PDFF).For reducing the acquisition time and obtaining PDFF simultaneously,this work proposed spine QSM sequence using multi-echo gradient-echo with two repetition times.In the proposed spine QSM sequence,out-phase echoes were acquired in the first repetition time,and in-phase(IP)echoes were acquired in the second repetition time.The field map and R2*calculated from IP echoes were served as R2*-IDEAL initialization for accurate field map estimation and water-fat separation(referred to as IP method).The results on both simulation and in vivo data demonstrated that the proposed IP method outperformed VARPRP-GC,SPURS,and Zero methods in field map estimation and water-fat separation.The proposed spine QSM method required about 5 minutes for data acquisition and could quantify QSM and PDFF simultaneously.(3)QSM reconstruction method based on multiple frequency convolutional neural network using the characteristic of dipole kernel in k-space.Deep learning convolutional neural network has great potential in QSM reconstruction.The existing deep-learning-based QSM reconstuction method(QSMnet)utilized convolutional neural network to substitute the field-to-susceptibility dipole inversion step,while the training process of QSMnet did not take advantage of the characteristic of dipole kernel in k-space.QSM artifacts are mainly caused by the frequency in the magic angle,where the value of dipole kernel is equal to or close to zero.For expression the characteristic of dipole kernel in the convolutional neural network,this work proposed multiple frequency convolutional neural network method(MF-QSMnet)that utilized the characteristic of dipole kernel in k-space for QSM reconstruction.Compare to QSM reconstructed by QSMnet,QSM reconstructed by the proposed MF-QSMnet was more accurate and had less artifacts.