DNN-based NMR Artifact Correction Method
Date
2024
Authors
Contributor
Advisor
Department
Instructor
Depositor
Speaker
Researcher
Consultant
Interviewer
Narrator
Transcriber
Annotator
Journal Title
Journal ISSN
Volume Title
Publisher
Volume
Number/Issue
Starting Page
Ending Page
Alternative Title
Abstract
Nuclear Magnetic Resonance (NMR) spectroscopy, a pivotal analytical technique in physics, chemistry, biology, and medicine, is frequently challenged by spectral distortions arising from various artifacts. These distortions can significantly hinder the accurate interpretation of NMR data, necessitating advanced correction methods. Thus, there have been many analytical methods to correct NMR artifacts, such as baseline distortion correction. This work introduces a deep learning-based approach for correcting NMR artifacts with bidirectional recurrent neural networks (BRNN) and convolutional neural networks (CNN). By simulating different types of datasets and applying these neural network architectures, this work demonstrates remarkable success in mitigating persistent NMR artifacts, mainly baseline and phase shift distortions. The proposed model was evaluated by calculating the accuracy of peak integration on real-world NMR spectra corrected by both the model and the conventional methods. The symmetricity of peaks from artifact-corrected spectra on actual NMR samples was calculated. Then, symmetricity was used to numerically assess the peak amplitude values for both distorted and symmetric peaks and calculate the accuracy of both correction methods.
Compared to traditional analytical methods, the proposed deep learning approach offers alternative ways to deal with NMR artifacts. Since the DNN approach is an automated NMR artifact correction, by applying consistent algorithms to identify and address artifacts, automated methods eliminate the potential for human bias. For a new type of NMR samples where the nature of baseline and phase shift distortion artifacts are different from previously known artifacts, researchers can focus on generating large sets of such data instead of developing computational methods to correct the artifacts. In order to facilitate such an approach, the model architecture is tested for memory efficiency for GPUs that are easily obtainable at the time of this writing, which allows training on GPUs with limited VRAM, unlike other DNN models designed for tasks of a similar nature.
Description
Keywords
Physics, Computational Physics, Deep Learning, NMR
Citation
Extent
186 pages
Format
Geographic Location
Time Period
Related To
Related To (URI)
Table of Contents
Rights
All UHM dissertations and theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission from the copyright owner.
Rights Holder
Local Contexts
Collections
Email libraryada-l@lists.hawaii.edu if you need this content in ADA-compliant format.