Hosting laboratory: Centre de Résonance Magnétique Biologique et Médicale (CRMBM),
Aix Marseille Univ, CNRS UMR 7339, Marseille
PhD supervisor: Guillaume Duhamel
PhD co-supervisor: Andreea Hertanu
Collaborators: Institute of NeuroPhysiopathology (INP), Marseille
Keywords: Magnetic Resonance Imaging, biophysical modelling, Monte Carlo simulations,
inverse problem solving, white matter microstructure
2026_ED352_Thesis_Proposal
PhD thesis description:
Magnetic Resonance Imaging (MRI) noninvasively probes biological tissues by exploiting the
nuclear magnetic resonance of hydrogen spins in water and macromolecules (e.g., lipids and
proteins). The measured MR signal arises from spin dynamics governed by relaxation,
molecular diffusion, and chemical exchange, all of which are modulated by the underlying
microenvironment. Quantitative MRI (qMRI), combined with biophysical modeling of the MR
signal, enables the decomposition of this aggregate signal into distinct proton populations by
representing spins with similar relaxation, diffusion, or exchange properties as separate
compartments1. In this framework, each compartment corresponds to a specific
microstructural environment (e.g., axonal, extra-axonal, or myelin). Estimating biophysical
properties through quantitative parameters such as relaxation times, diffusivity coefficients,
and exchange rates provides indirect access to tissue microstructure and composition,
offering a valuable approach to characterize and monitor tissue state in both health and
disease. In particular, the estimation of compartmental volume fractions serves as an
important indicator of the relative density of underlying microstructural constituents, such as
axons, myelin, and glial cells.
State-of-the-art biophysical models of white matter have provided important insights into
tissue microstructure and organization. However, they predominantly rely on single-contrast
mechanisms, probing parameters of relaxation, diffusion, or chemical exchange in isolation
through dedicated MR acquisition sequences that selectively sensitize the signal to one of
these processes. As a result, each class of models captures only a partial view of the
underlying microstructure and requires strong simplifying assumptions about tissue
composition and compartmental interactions. These simplifications lead to parameter
degeneracy and ultimately limit the specificity of the inferred microstructural features. For
instance, relaxation-based approaches typically model white matter as two water pools, often
associated with myelin water and the remaining tissue water2. Diffusion-based models, in
contrast, distinguish between intra-axonal and extra-axonal compartments but remain largely
insensitive to myelin-associated water3. Similarly, chemical exchange-based methods probe
interactions between bulk water and macromolecules, without explicitly separating distinct
microstructural compartments4
.
In the PHENIQS team (Physics and digital technologies for quantitative imaging of the central
nervous system) in CRMBM, we work on the development of a novel MRI acquisition
sequence that simultaneously sensitizes the MR signal to relaxation, diffusion, and chemical
exchange mechanisms. The objective of this PhD project is to design and validate a unified
biophysical model of white matter that jointly accounts for these mechanisms, providing a
more realistic description of tissue microstructure while reducing the biases inherent to single-
contrast approaches. The biophysical model of white matter will comprise five compartments:
water proton populations in the intra-axonal, extra-axonal, and myelin-associated spaces,
and macromolecular proton pools in the extra-axonal and myelin compartments. The
biophysical model will be interpreted by a mathematical model that combines Bloch–
McConnell equations5 (for relaxation and chemical exchange) and Bloch–Torrey equations6
(for relaxation and diffusion) into a single formalism.
In practice, realistic 3D in silico models of white matter, together with the mathematical
framework describing MR signal behavior, will be used in an open-source Monte Carlo
simulator7 to simulate the MR signal produced by the newly developed MR acquisition
sequence. This approach allows a controlled and quantitative assessment of how the
sequence interacts with the tissue microstructure, prior to any in vivo experiments. The in-
silico models of white matter will be generated with varying compositions of myelin, axons,
and glial cells using open-source tools8 and will be assigned different biophysical properties
of relaxation, diffusivity and chemical exchange. A sensitivity analysis will be performed by
systematically varying the acquisition variables of the MRI sequence across wide ranges to
identify the optimal acquisition settings that provide sufficient sensitivity to discriminate the
five compartments defined in the unified biophysical model.
To extract meaningful tissue properties from experimentally measured MR signals, the
inverse problem must be solved. The relationship between the signals and the underlying
biophysical parameters is nonlinear and sensitive to noise, and the large number of
parameters to estimate (22 in total) requires robust strategies for numerical stability.
Extensive datasets will be acquired on a calibrated physical phantom that replicates the five
compartments described by the unified biophysical model. Acquisitions will be guided by the
sensitivity analysis without any time constraint and performed on the 7T preclinical MRI
scanner in the CRMBM. Data subsets and parameter groupings will be iteratively refined
based on their impact on identifiability and fitting performance, until the estimated biophysical
parameters converge toward the ground-truth values.
Translation for in vivo experiments on mice will be considered in the last part of this PhD
project. For this purpose, the acquisition protocol will be optimized using analyses such as
the Cramér–Rao lower bound, with the final goal of reducing the acquisition protocol to
maximum 2 hours. The final protocol will then be applied to healthy mice, and repeatability
analyses will quantify robustness to physiological and experimental variability. Validation will
be achieved by correlating compartment-specific MRI metrics with corresponding histological
markers for myelin, axons, and glia (performed by collaborators in the Institute of
NeuroPhysiopathology in Marseille).
The ideal candidate should have a strong background in physics, biomedical engineering, or
a related quantitative field, with solid knowledge of MRI physics and signal modeling.
Experience in quantitative MRI and skills in scientific programming (Julia, Python, MATLAB)
are expected. The candidate should be rigorous, autonomous, and comfortable working in
an interdisciplinary environment at the interface of physics, mathematics, and biomedical
applications. Familiarity with numerical modeling, simulations, or inverse problems is a plus.
Good communication skills in English are required.
The innovative nature of this research project will offer the candidate valuable opportunities
to publish in leading international journals in the field (e.g., Magnetic Resonance in Medicine,
Imaging Neuroscience, NMR in Biomedicine), and to present their work at major international
and national conferences in the field, including ISMRM, ESMRBM and SFRMBM.
References:
1. Weiskopf N, Edwards LJ, Helms G, Mohammadi S, Kirilina E. Quantitative magnetic
resonance imaging of brain anatomy and in vivo histology. Nat Rev Phys. 2021;3(8):570-588.
doi:10.1038/s42254-021-00326-1
2. Mackay A, Whittall K, Adler J, Li D, Paty D, Graeb D. In vivo visualization of myelin
water in brain by magnetic resonance. Magn Reson Med. 1994;31(6):673-677.
doi:10.1002/mrm.1910310614
3. Jelescu IO, Budde MD. Design and Validation of Diffusion MRI Models of White Matter.
Front Phys. 2017;5. doi:10.3389/fphy.2017.00061
4. Henkelman RM, Huang X, Xiang QS, Stanisz GJ, Swanson SD, Bronskill MJ.
Quantitative interpretation of magnetization transfer. Magn Reson Med. 1993;29(6):759-766.
doi:10.1002/mrm.1910290607
5. Bloch F. Nuclear Induction. Phys Rev. 1946;70(7-8):460-474.
doi:10.1103/PhysRev.70.460
6. Torrey HC. Bloch Equations with Diffusion Terms. Phys Rev. 1956;104(3):563-565.
doi:10.1103/PhysRev.104.563
7. Cottaar M, Zheng Z, Miller KL, Tendler BC, Jbabdi S. Multi-modal Monte Carlo MRI
simulator of tissue microstructure. Imaging Neurosci. 2026;4:IMAG.a.1177.
doi:10.1162/IMAG.a.1177
8. Nguyen-Duc J, Brammerloh M, Cherchali M, et al. CATERPillar: a flexible framework
for generating white matter numerical substrates with incorporated glial cells. Med Image
Anal. 2026;110:103946. doi:10.1016/j.media.2026.103946