Artificial Intelligence-Assisted Low-Field Benchtop NMR Spectroscopy: Analytical Applications, Challenges, and Perspectives

Low-field benchtop nuclear magnetic resonance (NMR) spectroscopy has emerged as an accessible analytical platform for rapid, routine, and application-oriented analysis. However, its broader analytical adoption remains constrained by intrinsic limitations, including reduced spectral resolution, severe signal overlap, and lower sensitivity compared with conventional high-field instruments. To address these limitations, artificial intelligence (AI), including machine learning and deep learning approaches, has increasingly been explored alongside conventional chemometric strategies to enhance information extraction from low-field spectral data. This review examines recent developments in AI-assisted benchtop NMR across three major application domains: classification and authentication, quantitative analysis, and spectral processing or automated interpretation. Current evidence suggests that classification and authentication currently represent the most mature application area, whereas quantitative analysis shows promising but often condition-dependent performance. In contrast, spectral reconstruction and automated interpretation remain comparatively early-stage and exploratory, despite their potential long-term relevance for addressing intrinsic information limitations. Key challenges, including limited dataset diversity, poor model transferability, validation pitfalls, limited interpretability, and the lack of benchmarking and standardized workflows, are critically discussed. Future progress will likely depend not only on advances in AI algorithms, but also on the development of robust, reproducible, and analytically meaningful workflows. Overall, AI-assisted benchtop NMR is evolving from proof-of-concept applications toward a more structured analytical framework for extracting chemically meaningful information from spectrally constrained low-field data.