M2/engineer internship with follow-up PhD, « Advanced methods of blockwise diffusion imaging for studying fetal cerebral development at the mesoscopic scale »

Subject

Advanced methods of blockwise diffusion imaging for studying fetal cerebral development at the mesoscopic scale

Summary

Thanks to ultra-high field (11.7 teslas) blockwise imaging of post-mortem brains, we are studying fetal brain development with a resolution never before achieved in diffusion imaging (200 μm isotropic, p-HCP project). The objective of this internship will be to implement the first analyses of this unique data, in order to obtain precise reconstructions of the fibre pathways using tractography. The segmented nature of these blockwise acquisitions will require the implementation and development of appropriate algorithms. A PhD is planned following this internship, with the aim of developing the methods for performing these analyses on the whole brains, and thus realize the first mesoscopic-scale atlas of fibre development in the fetal brain.

Candidate profile

Prepared degree: Master 2 or engineering degree
Strong background in image analysis and/or applied mathematics
Programming: Python, C++
Interest in developmental neuroscience
Interest in experimental work, notably on anatomical samples
Fluent English, written and spoken

Practical

Internship duration and period: 5–6 months duration in year 2026

Internship payment: according to INSERM rules

Location:

NeuroSpin, bâtiment 145

Allée des Neurosciences

CEA Paris-Saclay

91191 Gif-sur-Yvette CEDEX

Team: inDEV (imaging neurodevelopmental phenotypes)

Supervision: Yann Leprince (NeuroSpin/UNIACT),

Co-supervision: Ivy Uszynski (NeuroSpin/BAOBAB/GAIA/Ginkgo)

Follow-up PhD: encouraged, subject to funding availability and candidate recruitment.

Background

The human brain undergoes extensive development during the second half of gestation. Key processes such as neurogenesis, neuronal migration, and axonal growth take place; transient structures form and disappear, such as proliferative zones and the subplate. These complex processes have been mostly studied with histology based on stained bidimensional sections of tissue, providing very detailed microscopic information on the tissue microstructure but relatively sparse spatial coverage [1]. More recently, magnetic resonance imaging (MRI) of prematurely born infants and in utero MRI have enabled whole-brain imaging at these ages, albeit with a spatial resolution limited to the coarse millimetric scale [2,3]. To bridge the gap between these scales, we propose to leverage an imaging technique recently developed in NeuroSpin: mesoscopic whole-brain imaging of ex vivo fetal brains using blockwise ultra-high-field MRI [4].

Full-brain mesoscopic imaging in humans is a task that has only been approached by ultra-high-field MRI and recently 3D histology [5], but both techniques have been limited to brains younger than 20 PCW (post-conceptional weeks) due to size constraints [6]. Indeed, only small-bore preclinical MRI scanners can provide the powerful gradients required for imaging the microstructure of the brain as well as the slot availability needed for extensive protocols. An original technique has been developed in NeuroSpin to overcome this limitation, consisting in blockwise imaging with a small-bore 11.7-tesla scanner followed by in silico reconstruction of whole-brain images. It was first developed to image an entire adult brain [7], and we recently adapted the method successfully to image the fetal brain [4] (ANR p-HCP, ANR-21-CE37-0029). This technique allows us to acquire multiparametric data at 100–200 μm isotropic resolution: quantitative maps of the T1, T2, and T2* relaxation times, and multi-shell high-angular-resolution diffusion imaging (b = 1500, 4500, 8000 s/mm² with 25, 60, and 90 directions respectively) at 200 μm isotropic resolution. As of the end of 2025, the data has already been acquired for six brains, and processed and published for three of them [8] (see Figure 1).

As of this first data release, the diffusion data has been processed separately for each shell, with the classical diffusion tensor imaging (DTI) model. This first step provides some general insight on the brain microstructure, but does not allow an accurate reconstruction of fibre bundle trajectories using streamline tractography. It also yields microstructural markers with poor specificity, compared to more advanced models such as NODDI (neurite orientation dispersion and density imaging) [9] or MSMT-CSD (multi-shell multi-tissue constrained spherical deconvolution) [10]. The reason for omitting these models is linked to the blockwise nature of the acquisitions: reconstructing whole-brain diffusion data is a challenge, particularly regarding the signal discontinuity at the border between blocks. Still, on the 18 PCW brain which was imaged as a whole, a proof-of-concept tractography and fitting of the NODDI model have been done successfully (see Figures 2 to 4), demonstrating the adequacy of our acquisition protocol.

PhD project

The PhD student will be tasked with developing and benchmarking methods for performing robust tractography on the blockwise data, in close partnership (co-supervision) with the Ginkgo team, which has leading expertise in diffusion imaging methods. To overcome the difficulty posed by signal discontinuity at the border between blocks, several approaches are envisaged: one approach consists in resampling the orientation distribution functions (ODFs) in a whole-brain space, and developing robust tractography algorithms that could skip over the imperfect boundary between blocks. Another approach could consist in performing the tractography in a blockwise manner with standard algorithms, and joining the tractograms at block boundaries. Gap-filling methods inspired by image inpainting could also be used to reconstruct realistic data where it is missing or corrupted at the boundary between blocks. This work is expected to yield the publication of a methods paper describing the algorithmic approaches and their benchmarking, as well as publication of the source code for the methods themselves.

The PhD student will also participate in the experimental work, i.e. the acquisition and semi-automatic reconstruction of new brains. Seven additional typical specimens are envisaged along the duration of the PhD, strengthening the coverage of the developmental timeline, which is essential given the rapid changes undergone by the brain during the second half of gestation. The addition of tractography and additional specimens will lead to a new public data release, and an associated data descriptor paper. Based on these results, publication of a neuroscientific paper describing aspects of the typical development of fibre architecture will be considered.

Finally, a proof-of-concept application to pathology will be made, aiming to image 3 brains with abnormal development of the fibre architecture (agenesis of the corpus callosum). A whole-brain description of the fibre architecture at the mesoscopic scale has never been shown in this malformative condition, and should lead to a publication, possibly in association with a clinician.