Single-Shot Quantitative Magnetic Resonance Imaging Based On Deep Learning

Magnetic resonance imaging has been widely used clinically because of its remarkable advantages.Quantitative imaging of magnetic resonance parameters can remove influences that are not related to tissue properties(eg,in different instruments,experimental sequences,sequence parameters,and experimental environments),and the quantitative properties of tissues,so MR parameters imaging gaining increasing attention.However,magnetic resonance imaging often results in long data acquisition time,and patient's conscious autonomous movement or unconscious involuntary movement can cause obvious artifacts.To address this issue,this paper presents a single-shot quantitative magnetic resonance mapping method based on overlapping echoes,including two overlapping echoes for quantified T2 mapping and four overlapping echoes for quantitative T2*mapping.Using an end-to-end neural network,quantitative,real-time,and motion-insensitive MR parameters mapping are acquired.For quantitative T2 imaging of two overlapping echoes,using two small-angle pulses generating two echo signals of different echo times,and then using an end-to-end neural network reconstruct the image after the excitation sequence.In this paper,the proposed method was verified by experiments.Experimental results were given on the human brain and quantitative analysis was performed.The results show that the proposed method can achieve real-time and high-quality imaging,and then we perform the sensitivity analysis of the movement.The results show that the proposed method is insensitive to motion and can be dynamically and quantitatively imaged.Finally,some comparative analysis experiments are performed to analyze the effects of different hyper-parameters on our model.The results show that the proposed method is less affected by hyper-parameters and has better robustness.We have also extended OLED technology to single-shot quantitative T2*imaging.For a long parameter value in the ROI(region of interest),the effect of the OLED reconstruction of two overlapping echoes may be poor.For longer parameter values,multiple echoes of the OLED image are more robustness.Therefore,we propose four overlapping echoes for OLED quantitative T2*imaging in order to expect a greater range of parameter values to be measured.The method uses four small-angle pulses generating four echo signals of different echo times,and then using an end-to-end neural network to reconstruct the images excited by the excitation sequences.In this paper,two models,RESNET model and U-shaped network model are proposed to solve the issue from different perspectives,and the reconstruction and quantitative analysis are performed on the clinical human brain.The method proposed in this paper promotes the development of single-scan multi-parameter magnetic resonance imaging and explores the potential for deep learning in complex MRI sequences.