Retaining both discrete and smooth features in 1D and 2D NMR relaxation and diffusion experiments

Abstract

A new method of regularization of 1D and 2D NMR relaxation and diffusion experiments is proposed and a robust algorithm for its implementation is introduced. The new form of regularization, termed the Modified Total Generalized Variation (MTGV) regularization, offers a compromise between distinguishing discrete and smooth features in the reconstructed distributions. The method is compared to the conventional method of Tikhonov regularization and the recently proposed method of L regularization, when applied to simulated data of 1D spin-lattice relaxation, T, 1D spin-spin relaxation, T, and 2D T-T NMR experiments. A range of simulated distributions composed of two lognormally distributed peaks were studied. The distributions differed with regard to the variance of the peaks, which were designed to investigate a range of distributions containing only discrete, only smooth or both features in the same distribution. Three different signal-to-noise ratios were studied: 2000, 200 and 20. A new metric is proposed to compare the distributions reconstructed from the different regularization methods with the true distributions. The metric is designed to penalise reconstructed distributions which show artefact peaks. Based on this metric, MTGV regularization performs better than Tikhonov and L regularization in all cases except when the distribution is known to only comprise of discrete peaks, in which case L regularization is slightly more accurate than MTGV regularization.

Multidimensional Proton Nuclear Magnetic Resonance Relaxation Morphological and Chemical Spectrum Graphics for Monitoring and Characterization of Polyunsaturated Fatty‐Acid Oxidation

“…To maximize the amount of information that can be extracted from LF‐NMR signal generation, we adapted and used a PDCO solver with excellent reconstruction accuracy that outperformed other in‐house NMR instrument's mathematical reconstruction programs (Wiesman et al, ). Signal reconstruction is essentially an ill‐posed ILT problem (Bortolotti et al, ; Reci et al, ); this implies that similar relaxation curves and noise in the measurements can result in very different reconstruction spectra and/or a high signal‐to‐noise ratio (SNR). In previous reports (Berman et al, ; Campisi‐Pinto et al, ; Wiesman et al, ), it was demonstrated that if an l 2 regularization term is added to the objective function of a common regularization, as seen in Eq.…”

1H LF‐NMR Energy Relaxation Time Characterization of the Chemical and Morphological Structure of PUFA‐Rich Linseed Oil During Oxidation With and Without Antioxidants

“…NMR signal reconstruction is an ill‐posed inverse Laplace transform (ILT) problem . Therefore, very similar energy relaxation time curves and/or a relatively low noise level in the measurements can result in very different reconstructed spectra and/or signal to noise ratio (SNR).…”

Analysis of Multicomponent Exponential Magnetic Resonance Relaxation Data: Automatable Parameterization and Interpretable Display Protocols

“…Many protocols are available, including imposition of non-negativity and the use of Tikhonov regularization. Developing and applying more advanced regularization methods that apply in specific situations, for example when the underlying distributions have both narrow and broad components (Borgia et al, 1998) (Reci et al, 2017), remains an active aspect of research in multi-component exponential relaxation data analysis.…”