Analysis of 1D NMR spectra with 2D image processing techniques

Abstract

Rapid advances in NMR spectroscopy instrumentation demand powerful analysis techniques to be implemented to cope with the development. These include extracting patterns from the data acquired, such as images and 1D spectra. Furthermore, deep learning models have shown to be very powerful in extracting patterns from such data. However, current analysis techniques include several manual steps to label the data, which are usually time-consuming and crucially dependent on expert knowledge. Therefore, this paper aims to demonstrate the applicability of neural networks for extracting different spectral structures and specially, investigating an analysis approach, inherited from 2D image processing, to determine its usefulness in feature extraction from 1D spectroscopy data. The approach utilizes bounding-box algorithms, same as in image recognition, to recognize patterns in 1D spectra.