AI-enhanced multi-parametric quantitative blood oxygenation level dependent magnetic resonance imaging of oxygen metabolism for clinical and neuroscientific application

• Applicants: Professorin Dr. Christine Preibisch; Professor Dr. Valentin Riedl; Professorin Dr. Julia Schnabel
• Subject Area: Radiology ; Medical Physics, Biomedical Technology
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 563563598

The cerebral metabolic rate of oxygen (CMRO2) is an important indicator of brain health and neuronal activity because the brain’s energy demand is ultimately covered by oxidation of glucose. Thus, in a clinical setting, assessment of oxygen metabolism can yield critical information, e.g., regarding brain tumor malignancy and severity of cerebrovascular diseases. In neuroscientific research, however, oxygen metabolism is considered as an important proxy of neuronal activity. While [15O] positron emission tomography (PET) is considered the gold standard for investigating the oxygen metabolism (Fan et al. 2020), magnetic resonance imaging (MRI) is more accessible and thus increasingly used to assess CMRO2 in healthy and diseased brain. Multi-parametric quantitative blood oxygenation level dependent (mq-BOLD) magnetic resonance imaging (MRI) is particularly robust and accessible. The mq-BOLD variant established in our lab has been successfully applied in brain tumors, stroke, carotid artery stenosis and neuroscientific studies. However, the required separate measurements of transverse relaxation times T2 and T2* as well as cerebral blood volume (CBV) and flow (CBF) still take about 15 min (pure acquisition time). Together with the need for contrast agent application for CBV measurement, this hinders more widespread application. In addition, as of now, the technique is susceptible to noise and artifacts, mainly due to magnetic background fields, subject motion as well as non-oxygenation-related susceptibility perturbations in white and deep grey matter. Further, there are a number of issues unaccounted for, importantly, the influence of intravascular signal contributions, diffusion, and multi-compartment tissue structure. Regarding recent progress in accelerated MRI and innovative computational methods, we hypothesize that mq-BOLD acquisition time can be considerably reduced, while addressing confounding influences and abolishing the need of contrast agent. The aim of the presented project is to advance the mq-BOLD technique for quantitative MRI-based CMRO2 mapping. In particular, we plan to employ novel accelerated and AI-enhanced MRI acquisition and reconstruction techniques in order to reduce measurement time, while – at the same time – reducing noise, artifacts and systematic bias. In addition, we pursue two different approaches to achieve contrast agent free CBV assessment, based on hyperoxia and arterial spin labeling (ASL), both supported by generative deep learning models. Overall, we strive to establish a robust and easily applicable tool that facilitates accurate CMRO2 mapping for routine neuroradiological diagnostics as well as neuroscientific research.

• DFG Programme: Research Grants
• International Connection: United Kingdom
• Co-Investigators: Privatdozent Dr. Carl Ganter; Dr. Stephan Kaczmarz; Professor Dr. Benedikt Wiestler
• Cooperation Partner: Professor Nicholas Blockley, Ph.D.