Description

This PhD position is part of the IMMENSE interdisciplinary project, which aims to develop new magnetic resonance methods combining spectroscopy, imaging and artificial intelligence to improve diagnostics in medicine and characterization tools in chemistry and materials science.

Magnetic resonance spectroscopy (NMR/MRS) provides unique information about molecular composition in fields ranging from brain metabolism to structural biology and materials science. However, spectral resolution strongly depends on the magnetic field strength. High-field instruments (e.g., 7 T MRI or ultra-high-field NMR) provide much higher spectral resolution than widely available lower-field systems.

This PhD project aims to explore a new paradigm:

Can AI reconstruct high-field spectral information from low-field measurements?

The goal is to develop foundation models for spectroscopy, capable of learning general representations of NMR signals and enhancing spectral resolution across different instruments and applications.

Research Objectives

The PhD will develop AI models for spectral representation learning and super-resolution applied to magnetic resonance spectroscopy.

Main objectives include:

Foundation models for spectroscopy: Develop deep learning models capable of learning general representations of NMR spectra across different magnetic fields and experimental conditions.

Spectral super-resolution: Train models to reconstruct high-resolution spectra from low-field acquisitions (e.g., 3 T → 7 T brain MR spectroscopy).

Cross-domain generalization: Investigate whether learned representations generalize across: brain metabolite spectroscopy, protein NMR, small-molecule spectroscopy, solid-state NMR for materials.

Spectroscopy datasets and benchmarks: Contribute to building standardized datasets of paired low-field and high-field spectra for training and evaluation.