Single-shot Quantitative Magnetic Resonance Imaging Based On Multiple Overlapping-echo Acquisition And Deep Learning

Quantitative magnetic resonance imaging(MRI)is of great value to both clinical diagnosis and scientific research.However,most MRI remain qualitative,especially dynamic MRI,because repeated sampling with variable weighting parameter makes quantitative imaging time consuming and sensitive to motion artifacts.As popular techniques to accelerate MR acquisition,compress sensing(CS)and paralleling imaging have been applied in quantitative parameter mappings.CS accelerates MR acquisition by sub-sampling MRI data and reconstructs complete information by utilizing sparsity prior of image,while paralleling imaging makes use of coil sensitivity maps of phased-array coils.However,for quantitative parameter mapping,CS acceleration or paralleling imaging acceleration is usually limited by the demand on evolution time to produce enough contrast information and remains unavailable for dynamic imaging.Recently,a new imaging acceleration method called overlapping-echo detachment(OLED)planar imaging is proposed.This method has been successfully applied to single-shot T2 mapping and single-shot diffusion mapping.Different from CS method,OLED method accelerates MRI acquisition by synchronously acquiring two echo signals with different parameter weighting which are overlapped in the k-space.These signals are separated afterwards by an optimization-based method based on structure similarity and sparsity prior,or are directly reconstructed through end-to-end mapping via deep neural network.In this thesis,we extend the OLED method to multiple overlapping-echo acquisition(called MOLED)and establish corresponding reconstruction algorithm based on deep neural network.The new method is applied to quantitative magnetic resonance T2 mapping and multi-parameter mapping.The main content of the thesis includes follows:1.Firstly,the principle of quantitative T2 imaging and conventional methods for quantitative T2 mapping are briefly introduced.Then,the OLED imaging sequence,imaging principle,convolution neural network and two reconstruction algorithms are introduced in detail,including every RF pulse and gradient in the pulse sequence,the expression of the overlapping echo signal in k-space,the optimal echo signal separation algorithm based on the structure similarity and the sparse prior,and the end-to-end reconstruction algorithm based on the deep neural network.2.A single-shot quantitative magnetic resonance T2 imaging sequence based on MOLED and its corresponding reconstruction algorithm based on deep neural network are proposed.Quantitative T2 imaging has great value in clinic,such as Parkinson's early diagnosis and cerebellar cortex research.We use the MOLED sequence to obtain multiple echo signals with different T2 weighting in a single shot,and reconstruct T2 mapping through deep neural network.Experimental results show that this method can not only achieve more accurate quantitative T2 mapping than previous methods,but also has strong robustness against non-ideal Bi field.3.A single-shot quantitative multi-parameter magnetic resonance imaging sequence based on MOLED and its corresponding reconstruction algorithm based on deep neural network are proposed.This method can acquire two k-space data with multiple echo signals in a single shot.Through proper reconstruction,T2 mapping and proton density image can be obtained at the same time.Simulation and phantom experimental results verify the feasibility of this method.