JTF-Net: A joint time-frequency domain deep learning network for accurate reconstruction and reference-free assessment of NMR Spectra

High-quality nuclear magnetic resonance (NMR) spectra can be rapidly acquired by combining non-uniform sampling techniques (NUS) with reconstruction algorithms. However, current deep learning (DL) methods are constrained to single-domain reconstruction (either in the time domain or frequency domain), leading to limitations like peak loss and artifact peaks, ultimately failing to achieve optimal performance. More importantly, the lack of fully sampled spectra presents challenges in assessing the quality of reconstructed spectra, making it difficult, even impossible, to determine their quality. In this study, we propose a joint Time-Frequency domain deep learning network, referred to as JTF-Net. JTF-Net effectively combines time domain and frequency domain features, exhibiting better reconstruction performance on protein spectra across various dimensions compared to traditional algorithms and single-domain DL methods. As the most significant breakthrough of this study, JTF-Net established a confidence lattice, and based on this confidence lattice, we propose the first reference-free quality assessment metric in the field of NMR spectrum reconstruction, denoted as Lattice Confidence Rate (LCR). LCR is a pioneering achievement, capable of evaluating the quality of reconstructed NMR spectra without the fully sampled spectra, making it more suitable for practical applications.