AA5 – Variational Problems in Data-Driven Applications


AA5-7 (was EF3-12)

Integrated Learning and Variational Methods for Quantitative Dynamic Imaging

Project Heads

Michael Hintermüller, Christoph Kolbitsch, Tobias Schäffter

Project Members

Moritz Flaschel

Project Duration

01.01.2022 − 31.12.2023

Located at



Emerging methods for clinical imaging aim to identify quantitative tissue parameter maps rather than qualitative gray scale images to boost the amount of information available for medical diagnoses.

Figure 1: Tissue parameter map obtained through quantitative magnetic resonance imaging.

In this project, a variational framework for inferring quantitative parameter maps directly from undersampled magnetic resonance imaging data in the k-space is proposed and investigated. Special emphasis lies on data-driven corrections to the possibly oversimplified physical models that constrain the inverse problem and on machine learning optimal regularization terms. The developed methods are integrated in a dynamic imaging framework with motion correction to deliver spatially and temporally varying parameter maps relevant for clinical applications like cardiac magnetic resonance imaging.

External Website

Selected Publications