Artificial intelligence is becoming a key software component for faster, robust, and quantitative magnetic resonance imaging.
Modern MR methods generate increasingly complex data. Advanced acquisition schemes, non-Cartesian trajectories, spectroscopic imaging, and dynamic metabolic imaging require sophisticated reconstruction, correction, and analysis methods. In our group, we develop artificial intelligence methods that transform MR data into reliable anatomical, functional, and metabolic information.
Our work focuses on the integration of deep learning with MR physics. Rather than treating AI as a black box, we aim to combine data-driven methods with physical models of MR signal formation, acquisition, reconstruction, and spectral quantification.
- We develop AI-based reconstruction methods to accelerate MR spectroscopic imaging and improve the quality of non-Cartesian data.
- Deep learning approaches are used to enhance spatial resolution, reduce acquisition time, and improve the robustness of metabolic imaging.
- Physics-informed models enable fast and reliable spectral fitting, allowing quantitative metabolite mapping from complex MR spectroscopy data.
- AI-based correction methods are developed to predict and compensate for motion-related changes, including dynamic B0 field variations at ultra-high field MRI.
- For deuterium metabolic imaging and other emerging MR techniques, we develop software tools that support fast reconstruction, denoising, fitting, and quantitative analysis.
- By combining artificial intelligence, MR physics, and advanced software development, we aim to create new measurement methods that make magnetic resonance imaging faster, more precise, and more clinically useful.
Unsere Entwicklungen
Artificial intelligence can be used directly during the acquisition process to make MR measurements more robust and adaptive. In our research, we develop AI-based methods for detecting and correcting motion-related effects, predicting dynamic B0 field changes, and improving the stability of MR measurements during scanning.
These approaches aim to reduce artefacts already at the measurement stage and to support future acquisition strategies that can react to physiological motion, subject movement, and temporal field instabilities.
AI methods play an important role in transforming raw MR data into high-quality anatomical, metabolic, and quantitative images. We develop deep learning approaches for reconstruction, signal separation, nuisance signal removal, and spectral fitting.
This includes methods for non-Cartesian MRSI reconstruction, water and lipid signal identification, accelerated neurometabolic imaging, and physics-informed spectral fitting. By combining neural networks with knowledge about MR signal formation, we aim to improve image quality, reduce scan time, and enable reliable metabolite mapping.
After image reconstruction, AI can support the extraction of clinically and scientifically meaningful information from complex MR datasets. Our work includes AI-based post-processing methods for quality assurance, automated evaluation, and quantitative image analysis.
We are particularly interested in robust pipelines that assess data quality, support reproducible analysis, and extract information from multi-contrast MRI. This includes applications such as multi-contrast segmentation in multiple sclerosis, automated quality control, and advanced analysis tools for large-scale MR studies.