Research On Magnetic Resonance Fingerprinting Imaging Technology And Multi-parameter Quantization Method

Magnetic resonance imaging is an important medical imaging technique,which has been widely used in clinical diagnosis.It can distinguish human soft tissue better,does not cause ionizing radiation to patients,and can obtain a variety of weighted images and other advantages.However,traditional magnetic resonance imaging is qualitative imaging.Doctors need to diagnose diseases relying on experience that cannot meet the needs of current precision medical treatment.Magnetic resonance fingerprinting(MRF),as a new quantitative magnetic resonance imaging technology,has the ability to obtain a variety of human tissue parameters through one data acquisition simultaneously,greatly improving the imaging speed and improving the impact of noise on image quality.However,the current research work on MRF is still in its infancy,and the parameter quantization is not accurate,especially the quantification of T2 parameters.In order to improve the image quality of MRF and raise the parameter quantization efficiency,we conduct the work from three aspects: data acquisition,dictionary design and generation and parameter quantization algorithm.First,the principle of MRF is studied.The MRF framework composes of three parts: data acquisition,dictionary design and generation,and parameter quantization algorithm.First,data acquisition was studied.The unique pulse sequence in MRF is studied,and the pseudo-random combination of FA and TR is designed.Then,the pseudo-random data acquisition method is applied to collect multi-frame data for the same fault plane.Then the dictionary design and generation and dictionary compression was studied.When designing the dictionary,we take three parameters of T1,T2,and PD into account.In addition,we also consider the physical quantity of B0(non-uniformity of magnetic field).Considering the impact of dictionary size on computational complexity and memory consumption,we compress the dictionary using singular value decomposition,thus obtaining a compressed dictionary.Finally,the basic principles of the traditional parameter quantization algorithm are studied in detail.Then,the parameter quantization algorithm based on iterative method is studied.Firstly,the MRF data acquisition model is established.The idea of compressed sensing is used to classify the high undersampling problem into constrained convex optimization problem.Then the gradient projection algorithm is applied to solve the constrained convex optimization problem.The gradient update is applied to acquire gradually multiple frame high-quality MRF spatial domain images(including fingerprint signals)in the iterative process,and then projection is performed on the more accurate fingerprint signals to find the best matching entries in the dictionary pixel by pixel.The difference between the Bloch response iterative projection method and the parameter quantization method based on the cover tree and the approximation nearest neighbor search is the projection step in the iterative process(ie,finding the best match corresponding to the fingerprinting),the accurate nearest zero search used by the former.The latter replaces the original nearest neighbor search with an inexact(1+?)cover tree to approximate the nearest neighbor search,and takes the excellent data structure of the cover tree.Finally,the improvement method based on cover tree and approximation nearest neighbor is proposed,mainly including the combination of {TR,FA},the combination of {T1,T2},the EPI sampling trajectory,and the search step in the iterative process.Then,the algorithm is implemented with the shangguocan model data and the Brainweb data.The bSSFP pulse sequence was used to perform 16 x and 32 x undersampling with a uniform EPI trajectory.Several pattern matching algorithms are compared,including direct matching method,Bloch response iterative projection method,iterative projection method based on cover tree and approximation nearest neighbor search method,and improved method.Besides the absolute error image is counted in the experimental results.Quantitative evaluation such as mean absolute error(MAE),normalized root mean square error(RMSE),and running time were calculated.The results suggest that the improved method is better than the traditional method,which can greatly improve the quality of the MRF multi-parameter image(T1,T2,B0,PD),and was realized easily within an acceptable time.In addition,the improved algorithm is not sensitive to random additive noise.There are still many technical difficulties in the field of MRF imaging that need to be overcome.The improved parameter quantization method greatly improves the image quality of T1 and T2 while improving the quantization efficiency,which provides certain technical support for the clinical application of MRF.