Signal Processing Methods For Nuclear Magnetic Resonance Based On Deep Learning

High quality magnetic resonance spectrum signals can effectively and accurately obtain abundant molecular structure information.However,there are many interference factors,such as phase distortion and noise,which affect the quality of the spectra.In order to obtain the ideal absorption line spectrum,it is necessary to preprocess the acquired magnetic resonance signal.The automatic pretreatment can avoid timeconsuming and laborious manual operation and obtain the ideal absorption line spectrum effectively.On the other hand,in the case that the analyzed sample is too complex,the obtained spectral map often has the problems of overlapping spectral peaks and low resolution.Therefore,how to reasonably simplify the nuclear magnetic spectrum signal and improve the resolution is one of the current research hotspots.In recent years,with the continuous development of deep learning,its applications in image,speech and natural language processing are becoming more and more mature,showing its strong learning ability.In this paper,the deep learning technology is used to automatically correct the phase distortion in the magnetic resonance signal and decompose the J coupling for the purpose of simplifying the spectrum.The main research content is divided into the following three parts:1.In this paper,a phase correction method based on deep learning of magnetic resonance spectrum is proposed.The phase correction and denoising of magnetic resonance spectrum signals with phase distortion and noise are processed by network learning.2.In order to achieve the true end-to-end study,we try to explore on the basis of phase correction involving domain transform method,the input for the phase distortion and the FID signals with noise,using deep learning network has realized the timefrequency transform,phase correction,denoising processing after get ideal absorption line spectrum.3.A wideband uncoupling method based on deep learning is proposed.Through network learning decoupling and de-noising,the purified degree shift spectrum with lorentz line is obtained by de-pairing and de-noising the normal spectrum signal with J coupling and with noise.The validity range of the method is tested by simulating the data of various feature types.To sum up,this paper tries to use deep learning technology to put forward the above solutions to the problems of phase correction and pair removal,and test the effectiveness of the method in all aspects,and explore its effectiveness range.