Correction And Quantification Of Nuclear Magnetic Resonance Spectroscopy In Inhomogeneous Magnetic Fields

One of what affects the quality of NMR spectrum and its ability to be effectively utilized is the linewidth of the peak which is related to the spectral resolution,while linewidth or spectral resolution is highly dependent on the environment of magnetic field in which the sample is detected.As a technology that can non-invasively measure metabolite information of biological tissues,localized spectra have shown excellent results in clinical auxiliary diagnosis and biological research.However,in the experiments of biological tissues,the inhomogeneous magnetic fields also trouble the technicians.Although the non-uniformity of magnetic field environment can be mitigated by various hardware or experimental methods,remaining uniform at all time is unrealizable in usual.For instance,biological tissue may produce inhomogeneous magnetic fields due to variance in the inherent magnetic susceptibility.Up to now,in the research of resisting the effect caused by inhomogeneous magnetic fields,most use experimental methods,but the data post-processing ways are less effective and perfect.Here,considering the increasingly developed technology of data processing such as deep learning which has been successfully applied in many fields,a series of neural networks can be potentially combined with NMR to achieve automatic correction and quantification of the MRS in inhomogeneous magnetic fields.In the first part of this thesis,a method that uses neural network algorithm to obtain high-resolution 1H MRS spectra in inhomogeneous magnetic fields is proposed.A spectrum with lower resolution and lower SNR(signal-to-noise ratio)acquired in inhomogeneous magnetic fields was corrected to an ideal and readable one.Subsequently,simulated data and experimental data were tested respectively,and the evaluation indicators are discussed and analyzed as well.In the second part of this thesis,an end-to-end quantification method of 1H MRS in inhomogeneous magnetic fields is proposed.Based on the first research in the thesis,a transfer learning way is applied in this part to obtain precise quantification of metabolite concentration from a spectrum with low resolution and low SNR acquired in inhomogeneous magnetic fields.