不均匀磁场下高分辨率核磁共振波谱重建方法

Rich information of molecular structure and dynamics can be delivered by high resolution nuclear magnetic resonance (NMR) spectrum, to obtain which the spectrometer is required to provide highly homogeneous static (B0) fields. However, there are many circumstances, e.g. when magnetic susceptibility of sample is spatially inhomogeneous, where the high homogeneity of B0 is unobtainable, thus results in a decrease in resolution of the spectra. Applying spatial encoding in the pulse sequence, the space is divided into small voxels, and the NMR signal coming from each voxel is in relatively high homogeneous B0. Based on this feature of spatial encoding technology, this research project involves the following work: (1) To obtain high resolution spectra in inhomogeneous B0 fields, two typical spatial encoding pulse sequences will be investigated, where EPSI (echo planar spectroscopic imaging) module is applied for the field map imaging. The sources of noise within the spatial encoding NMR signal will be analyzed. Modify the two pulse sequences and apply advanced signal processing techniques to improve the signal to noise ratio and the spatial resolution. (2) Design the denoise algorithm to better serve the spatial encoding spectra. To suppress noise while keeping the weak signal components as much as possible, using the statistics of signal and noise in each voxel to increase the signal to noise ratio. (3) Develop an NMR signal decomposition method based on multiobjective optimization evolutionary algorithms, for the purpose of extracting the properties (i.e. frequencies, amplitudes, relaxation times, and initial phases) of the signal components, which will then be applied to the high resolution spectra reconstruction.

高分辨率核磁共振谱可提供丰富的分子结构、动力学等信息，然而也要求仪器提供在空间上高度均匀的磁场。在样品磁化率不均一等情况下磁场的均匀性很难保证，导致难以获得高分辨谱图。利用空间编码技术的空间解析能力，可以得到小体素内的核磁信号，而每一个体素内的核磁信号都具有高分辨率，基于此本项目拟开展如下方面研究：（1）为获取不均匀磁场场图检测，采用两种基于回波平面成像模块的空间编码序列，分析空间编码信号在采集过程中遇到的噪声的类型，改进序列并通过后期的信号处理提高信噪比。（2）设计适用于空间编码谱的消噪算法，在尽量不损失微弱信号的前提下，利用各体素内信号的统计特性消除噪声、提高信噪比。（3）开发一种多目标优化的演化算法用于信号分解，以提取信号的特征信息，并用于空间编码谱图的识别与不均匀度校正，从而实现高分辨率磁共振谱重建。

本项目针对磁共振波谱信号在不均匀磁场下频率分辨率低、信噪比低的问题，尝试从序列设计与信号处理两个方面提高频谱分辨率和信噪比。围绕这一目标，我们提出了一系列的方法，主要成果有：（1）为了提高频谱分辨率，我们改进了不均匀场下的纯化学位移谱技术，并设计了相应的谱图重建算法，得到了超高分辨率的谱图效果。（2）为了提高谱图的信噪比，我们设计了一种随机SVD分解结合软阈值的算法，可以在快速、有效地去除噪声的同时尽可能地保留信号的强度。（3）为了提高频谱分辨率，避免复杂样品产生信号之间的相互干扰，我们设计了盲源分离方法，以获取最少谐波组分的信号为目标，可以较好地解析出复杂的混合信号。（4）为了分析具有不同扩散系数的复杂样品信号，采用扩散排序磁共振谱，并设计了基于稀疏与低秩特性的反拉普拉斯重建算法以获取超高分辨率的扩散排序谱图。（5）我们还将深度学习引入核磁共振谱图的处理中，研究了基于深度学习的高分辨核磁图像重建方法，以及基于深度学习的核磁谱去噪及相位校正方法。