The holy grail of quantitative MRI (qMRI) is the accurate and precise quantification of clinically relevant biological tissue properties. Significant advances have been made in the rapid and reliable acquisition of the classical, MRI-physics-related quantities (T1, T2, R2*, PD, etc.). Despite this progress, the clinical relevance of qMRI has been limited, in part due to research demonstrating that qMRI quantities have limited specificity for the underlying biological tissue composition. It has been found that most qMRI quantities are co-dependent on multiple tissue properties, such as iron and myelin, making the interpretation of qMRI findings difficult in both clinical and research settings.
A relatively recent approach toward increasing qMRI’s specificity toward biological properties is the analytical combination of multiple qMRI quantities. However, combining competing signal contributions from iron and myelin with the involved non-linear signal models remains unachieved. Furthermore, current approaches require physical model parameters that are not known with sufficient precision in most cases and have to be estimated from external measurements. It is also not known if these parameters vary between subjects or if they are affected by the disease.
We are striving to overcome the limitations of current methods by developing physics-based machine-learning approaches. Our approach is to utilize customized neural network architectures that are specifically tailored to the physical signal constraints of the tissue in order to obtain tissue property maps that are more biologically accurate. We hope that this will help us to better understand the biological processes at work.
Publications
Self-calibrating χ-separation using single-subject (N=1) physics-constrained deep learning.
Benslimane I, Grabner G, Hametner S, Jochmann T, Zivadinov R, Schweser F
15th Annual Buffalo Neuroscience Research Day, Buffalo, NY, 2022
Benslimane I, Grabner G, Hametner S, Jochmann T, Zivadinov R, Schweser F
Joint Workshop on MR phase, magnetic susceptibility and electrical properties mapping, Lucca, Italy, 2022
Benslimane I, Grabner G, Hametner S, Jochmann T, Zivadinov R, Schweser F
Proc Intl Soc Mag Reson Med, 2022, p. 370
Benslimane I, Grabner G, Hametner S, Jochmann T, Zivadinov R, Schweser F
Artificial Intelligence in Clinical Imaging (AICI) Forum, Graz, Austria, 2022
Mir S, Salman F, Jakimovski D, McGranor C, Zivadinov R, Schweser F
Joint Workshop on MR phase, magnetic susceptibility and electrical properties mapping, Lucca, Italy, 2022
A Pipeline for the Determination of the Iron Microstructure Coefficient (IMC) from MGRE Data.
Mir S, Salman F, Schweser F
15th Annual Buffalo Neuroscience Research Day, Buffalo, NY, 2022
Benslimane I, Jochmann T, Zivadinov R, Schweser F
Proc Intl Soc Mag Reson Med, 2021, p. 330
Yanis Taege, J. Hagemeier, N. Bergsland, M. Dwyer, B. Weinstock-Guttman, R. Zivadinov, F. Schweser
NeuroImage, 2019
Non-invasive Investigation of the Compartmentalization of Iron in the Human Brain
F. Schweser, J. Sedlacik, A. Deistung, J. Reichenbach
2012
SIAMESE-TWINS: Quantitative Kartierung von Eisen und Myelin im menschlichen Gehirn
Ferdinand Schweser, A. Deistung, Bw Lehr, J. Reichenbach
2011
F. Schweser, A. Deistung, K. Sommer, J. Reichenbach
2011
SEMI-TWInS : Simultaneous Extraction of Myelin and Iron using a T 2 *-Weighted Imaging Sequence
F. Schweser, A. Deistung, B. W. Lehr, K. Sommer, J. Reichenbach
2010
Non-linear evolution of GRE phase as a means to investigate tissue microstructure
F. Schweser, A. Deistung, D. Güllmar, M. Atterbury, B. W. Lehr, K. Sommer, J. Reichenbach
2010