Reconstruction of nuclear magnetic resonance spectroscopy with deep learning

• Main Author: Zang, Jiayu
• Other Authors: Yap Kim Hui
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/155730
• Institution: Nanyang Technological University

Nuclear magnetic resonance (NMR) spectroscopy is widely used as an effective analytical tool for the analysis of chemical substances and proton structures. Multidimensional NMR spectra can provide more information about the structure of chemical substances compared to one-dimensional spectra, but as the dimensionality of the spectra increases, the experimental time required for the spectra also increases significantly. Non-uniform sampling (NUS) can shorten the experiment time by reducing the number of sampling points in the indirect dimension of the spectrogram. Also, for non-uniform sampling data, a suitable reconstruction algorithm is needed for fast reconstruction of the spectra. Since its formal introduction, the concept of deep learning (DL) has had a major impact on academic research and the computer industry, providing an unprecedented approach to analyzing and processing data and leading to great achievements in computer vision, medical imaging, natural language processing, and more. In this thesis, we explore the feasibility of using deep learning methods for the reconstruction of non-uniformly sampled NMR spectra. We employ a neural network that can be interpreted at the layer level, and each layer of this network corresponds to an iterative operation of the Iterative Shrinkage-Thresholding Algorithm. The experimental results show that our proposed network has better reconstruction performance compared with the traditional networks ResNet and DenseNet, which further validates the improvement of the interpretable network structure on the network performance.