Research On The Quantitative Method Based On The Dictionary - Free Magnetic Resonance Fingerprinting Technology

"Magnetic resonance fingerprinting" imaging technology is developed on the basis of magnetic resonance scanning.This method resolves the defect of MRI technology that the scanning time is too long,it can’t obtain quantized image and only one weighted image can be obtained at a time.MRF technology is a new imaging technology,completely different from MRI technology in terms of sequence design,data collection and data processing.However,the accuracy and efficiency of MRF parameter quantization is an urgent problem in this field.First,we investigate the principle of magnetic resonance fingerprinting technology and study the dictionary-based magnetic resonance fingerprinting imaging method.In the scanning process of MRF imaging technology,in order to obtain data with time and space irrelevancy,we set the flip Angle and repetition time with pseudo-random changes on the sequence.Nonuniform K spatial data with high undersampling can be obtained by radiofrequency pulse and gradient pulse.A set of reconstructed images can be obtained by Non-uniform Fourier transform of the acquire data.After the reconstruction,connect the signal of each voxel point in each reconstructed image can obtained the characteristic curve of voxel under this sequence.It named "magnetic resonance fingerprinting".Using this method requires building dictionary library.It is necessary to select the range and precision of corresponding parameters according to different parts and tissues,it determines the number of entries in a dictionary.Building a dictionary library requires that all parameters be combined with the same sequence used to collect data,and that the dictionary signal be simulated through the Bloch equation(a set of parameters combined to get a dictionary signal).Match the fingerprinting signal with all the entries in the dictionary database one by one to get the most similar dictionary signal.Tissue’s multiple parameters can be inferred from the parameter combination information when the dictionary is generated.Finally,the quantitative map can be obtained.We are also studied the influence of different sequences on the signal characteristics of the dictionary and the different dictionary matching algorithms.Then,we study the dictionary-free MRF imaging algorithm based on extended kalman filter(EKF-MRF).In the quantization of MRF parameters based on EKF method,we should establish the system dynamic equation at first,Then the estimated signal is tracked by EKF iteration.The process of calculating MRF parameters by EKF iterative method is mainly divided into two parts: prediction and update.In the prediction process,the state vector at the next moment(that is,the magnetization vector at the next moment and each parameter value)and the variance matrix at the next moment are predicted by the optimal estimation at the current moment.Then calculate the kalman gain.Then the prediction data(state vector and variance at the next moment)are updated by the minimum variance principle,and the state vector estimation and variance matrix of the optimal estimation are obtained.Through EKF iteration,the estimated value is close to the real value,and then the input parameters of the signal are reversely deduced and estimated,and the quantized parameter values are finally obtained.During the update process,EKF iteration is carried out continuously to make the estimated value close to the real value,so that the input parameters of the signal are reversely deduced and estimated,and the quantized parameter values are obtained.In the study of EKF-MRF method,observation values are obtained by imaging model simulating MRF acquisition process.In EKF process,we continuously adjust the value of process noise Q so that the value of Q becomes smaller and smaller in the iterative process.In other words,more and more weight is given to the predicted value,so that the estimated value converges faster and tends to the real value.The value of the observed noise R is adjusted to make the estimated value range converge more accurately.Changing the Q value and R value reduces the iteration times of the EKF process,thus reducing the time required by the algorithm.In addition,in the EKF iteration process,the first-order linearization times of the nonlinear system state equation are changed.By calculating the first-order linearization of the nonlinear system state equation only once,the computational burden of the algorithm is reduced and the computational efficiency of the algorithm is accelerated.Finally,we developed a software for the visualization of "magnetic resonance fingerprinting" parameter quantization method in MATLB environment.The software can select different sequences and required reconstruction models according to requirements and it can realize the visualization of more than one parameter quantization methods.We compare and analyze the results under different methods.The analysis results show that MRF parameter quantization technology based on EKF can no longer be affected by dictionary dependence.It does not need to build different dictionaries for different sequences and different parts of the organization,which avoids the waste of time and space caused by building dictionaries.Secondly,compared with the direct dictionary matching method,EKF algorithm not only greatly speeds up the operation speed,but also obtains the parameter quantization image with smaller error.