Deep learning in MR Spectroscopic Imaging

Magnetic resonance spectroscopic imaging (MRSI) is a powerful technique that is able to visualize the spatial distribution of biochemical compounds in the human brain. With a single MRSI examination multiple such biochemical compounds can be mapped. These biochemical compounds play fundamental roles in the functioning of the healthy brain as well as in brain disorders and are not accessible via conventional Magnetic resonance imaging methods. While there are many research applications for MRSI, the clinical application of MRSI lags behind considerably. The main reason is the lack of a simple and robust push- button MRSI tool that is fully integrated on MR scanners. The main obstacles are the time- consuming reconstruction process and the currently necessary manual quality assurance. Recent advances in artificial intelligence, specifically related to so-called deep neuronal networks, raise the hope that this barrier can soon be overcome. In this project, we aim to develop a push-button tool for non-invasive mapping of the brain`s biochemistry using deep learning based MRSI techniques. This should make MRSI more accessible for neuroscience and clinical use.

In this project we have successfully created an AI-supported method that lets MRI scanners produce clear "brain chemistry maps" quickly and reliably. We brought the whole process together into one easy workflow, so the results are available soon after the scan without complicated manual steps. By testing it across multiple sites, we showed it works consistently in different hospitals and research centers, making it ready for broader use in research and, in the future, patient care.