UNMR-Net: A Unified Deep Reconstruction Network for Multitask Compressed Sensing NMR Spectroscopy

Area of Science:

• Chemistry
• Life Sciences
• Artificial Intelligence

Background:

• Nuclear Magnetic Resonance (NMR) spectroscopy is crucial in chemistry and life sciences.
• High resolution in NMR is often limited by long acquisition times.
• Current deep learning (DL) models for NMR are often specialized and lack transparency.

Purpose of the Study:

• To develop a unified deep reconstruction network for multitask compressed sensing NMR spectroscopy.
• To improve the generalizability and interpretability of DL models in NMR data reconstruction.
• To accelerate NMR acquisition while maintaining spectral quality.

Main Methods:

• Unrolling a classical iterative shrinkage-thresholding algorithm into a deep network architecture.
• Developing a multitask framework adaptable to different NMR sparse reconstruction tasks.
• Combining model-driven reliability with data-driven expressivity.

Main Results:

• The unified network achieved superior reconstruction performance compared to specialized models.
• The model demonstrated adaptability by being applied to two different NMR sparse reconstruction tasks.
• Stage-wise visualization provided improved transparency into the reconstruction process.

Conclusions:

• The proposed model addresses adaptability and interpretability concerns in DL for NMR.
• This approach supports more reliable DL-assisted spectroscopic interpretation.
• The findings boost the potential of AI to expand NMR methodologies.