Internship Offer

Functional and topological data analysis in neuroimaging

Merci d’envoyer un CV ainsi que les coordonnées d’au moins une référence à :
functional data
topological data

Bertrand Michel

Aymeric Stamm


March 4, 2024

Internship proposal

Key elements

  • Title: Functional and topological data analysis in neuroimaging
  • Internship duration: 6 months
  • Research unit: Department of Mathematics Jean Leray, UMR CNRS 6629
  • Workplace: Department of Mathematics Jean Leray, UMR CNRS 6629, Nantes University, 2 chemin de la Houssinière, 44322 Nantes Cedex 3, France.
  • Salary : 669,90 euros per month (22 working days).
  • Funding agency: Agence Nationale de la Recherche (ANR) - Project PASTRAMI.

Keywords: functional data analysis, Wasserstein distance, optimal transport, space of parametric mixture models, personalized medicine, geometric inference, topological data analysis, hypothesis testing, graphical representation of hypothesis tests, kernel methods, clustering.

Ideal candidate profile

  • Master 2 or last year of engineering school with specialization in probability and statistics;
  • Good programming skills (R, Python, C/C++);
  • Good scientific English reading skills (capable of reading scientific papers written in English);
  • Enthusiastic about topics related to neuroimaging and generally speaking interdisciplinary topics;
  • Autonomous and rigorous.

Neuroimaging context

Traumatic brain injury (TBI)

Traumatic brain injury (TBI) represents 1.5 million hospital admissions in the European Union (EU) each year and approximately 160,000 cases/year in France. Their causes are numerous: accidents, contact sports, military (Bhattrai, Irimia, and Van Horn 2019), etc. It is a leading cause of injury-related death (57,000 TBI-related deaths each year in the EU) and disability, with a devastating impact on patients and their families (Majdan et al. 2016). It is therefore of paramount importance both from a clinical standpoint and from an ethical and societal standpoint to be able to accurately predict functional outcomes. The PASTRAMI project aims to develop a patient-specific prediction model for TBI functional outcomes using neuroimaging data.

Brain microstructure and structural connectivity

The human brain is a complex organ composed of approximately 86 billion neurons and 85 billion non-neuronal cells. The neurons are connected to each other through synapses, which are the key elements of the brain’s information processing system. The brain’s white matter (WM) is composed of axonal bundles that connect different regions of the brain and glial cells that support and protect the neurons. The axonal bundles are organized in a complex network of connections, the so-called connectome. Biological parameters of interest in the context of TBI include axonal density and mean diameter, axonal orientations, glial cell density and mean diameter and proportion of free water. In effect, severe TBI is often associated with axonal injury, which often translates into a decrease in axonal density and mean diameter, and an increase in the proportion of free water.

Diffusion MRI

Diffusion MRI is an imaging modality sensitive to constrained water diffusion in tissues. It is the only non-invasive imaging modality that can provide information about the brain’s microstructure and structural connectivity (Alexander et al. 2019). In particular, using an appropriate model of the diffusion process (Assaf et al. 2004), the above-mentioned biological parameters of interest can be estimated at each voxel of the brain. Once this information is estimated, it can be used, through the process of tractography, to provide a mathematical reconstruction of axonal bundles (Jeurissen et al. 2019). The result of all the preprocessing steps is a set of mathematical objects that jointly represent the brain’s microstructure and structural connectivity. These objects are high-dimensional and complex, and their analysis is a challenging task. In particular, the number of axonal bundles is not known a priori and varies from one individual to another. This makes the comparison of a patient’s brain to healthy brains difficult.

Internship objectives

Mathematically, we define a streamline as an ordered sequence of 3D points \(x_1, \dots, x_M\). Taking the view of functional data analysis, a streamline can be viewed as a parametrized curve \(x(s) \in \mathbb{R}^3\), \(s \in [0,1]\), with \(x(0) = x_1\) and \(x(1) = x_M\). A white matter fascicle is defined as a collection of streamlines connecting to regions of the brain and responsible for the communication between them with the aim of performing a specific function. In addition, at each sampled point of each streamline composing the fascicle, one can add the information about the local microstructure (local tissue composition: number of tissue populations and respective volumes of occupancy, axon density, axon diameter, glial cell diameter, etc.). This information is recovered via diffusion MRI by modeling and estimating constrained diffusion of water in tissue with impermeable membranes and specific geometries (e.g. cylinders or spheres) as a mixture of parametric probability distributions. If we denote by \(\mathcal{M}\) the set of such mixtures of parametrized distributions, we can now define a microstructure-augmented fascicle (MAF) as a collection of parametrized curves \(f_1(s), \dots, f_N(s) \in \mathbb{R}^3 \times \mathcal{M}\), \(s \in [0,1]\), with \(f_i(0) = (x_1^{(i)}, p_1^{(i)})\) and \(f_i(1) = (x_M^{(i)}, p_M^{(i)})\), where \(p_j^{(i)}\) is a mixture of parametric distributions describing diffusion of water (and thus, indirectly, microstructure) locally around the point \(x_j\) on the \(i\)-th streamline.

A MAF of primary importance is the corticospinal tract (CST) sometimes also referred to as the pyramidal tract (PyT), which is the main motor pathway in the human brain. It is responsible for voluntary movement and connects the motor cortex to the spinal cord. The CST is a well-known target for neurosurgical procedures and is often affected in TBI. The method that generates MAFs from diffusion MRI data is called tractography and outputs a number of streamlines that are supposed to represent the fascicle of interest. However, the output of tractography is highly variable and depends on the acquisition protocol, the tractography algorithm, and the post-processing steps. This variability makes it difficult to compare MAFs between patients and healthy controls. The goal of the internship is to develop a statistical model for comparing MAFs between a patient and a healthy population by building a adequate representation of the CST via representative streamlines in such a way that this representation be identifiable across subjects and robust to the variability of tractography.

Care will be taken to ensure that the representation is also interpretable in terms of the underlying microstructure. The candidate will therefore explore the existing literature on Wasserstein barycenters for mixture models (Peyré, Cuturi, et al. 2019; Delon and Desolneux 2020) and propose a method to build a representative MAF for the CST. The candidate will also explore the existing literature on the use of functional data analysis (FDA) or topological data analysis (TDA) for:

  • clustering streamlines: in FDA, we can use the SRF or SRVF representation of the streamlines and apply clustering methods to such data (Kurtek et al. 2012; Srivastava and Klassen 2016). In TDA, we can use the persistence diagrams of the streamlines and apply clustering methods to the persistence diagrams (Chazal and Michel 2021).
  • comparing the representative MAF to the MAFs of the healthy population which can be framed either as an hypothesis testing problem or as an anomaly detection problem. In both cases, it is of paramount importance not only to tell if there are anomalies but to localize them and to provide a measure of the severity of the anomalies.

The candidate will implement the proposed methods and apply them to a data set of diffusion MRI data from patients with TBI and healthy controls. The candidate will also participate in the writing of a scientific article and the preparation of a presentation of the results.

The internship might be extended to a PhD thesis.


Alexander, Daniel C, Tim B Dyrby, Markus Nilsson, and Hui Zhang. 2019. “Imaging Brain Microstructure with Diffusion MRI: Practicality and Applications.” NMR in Biomedicine 32 (4): e3841.
Assaf, Yaniv, Raisa Z Freidlin, Gustavo K Rohde, and Peter J Basser. 2004. “New Modeling and Experimental Framework to Characterize Hindered and Restricted Water Diffusion in Brain White Matter.” Magnetic Resonance in Medicine: An Official Journal of the International Society for Magnetic Resonance in Medicine 52 (5): 965–78.
Bhattrai, Avnish, Andrei Irimia, and John Darrell Van Horn. 2019. “Neuroimaging of Traumatic Brain Injury in Military Personnel: An Overview.” Journal of Clinical Neuroscience 70: 1–10.
Chazal, Frédéric, and Bertrand Michel. 2021. “An Introduction to Topological Data Analysis: Fundamental and Practical Aspects for Data Scientists.” Frontiers in Artificial Intelligence 4: 108.
Delon, Julie, and Agnes Desolneux. 2020. “A Wasserstein-Type Distance in the Space of Gaussian Mixture Models.” SIAM Journal on Imaging Sciences 13 (2): 936–70.
Jeurissen, Ben, Maxime Descoteaux, Susumu Mori, and Alexander Leemans. 2019. “Diffusion MRI Fiber Tractography of the Brain.” NMR in Biomedicine 32 (4): e3785.
Kurtek, Sebastian, Anuj Srivastava, Eric Klassen, and Zhaohua Ding. 2012. “Statistical Modeling of Curves Using Shapes and Related Features.” Journal of the American Statistical Association 107 (499): 1152–65.
Majdan, Marek, Dominika Plancikova, Alexandra Brazinova, Martin Rusnak, Daan Nieboer, Valery Feigin, and Andrew Maas. 2016. “Epidemiology of Traumatic Brain Injuries in Europe: A Cross-Sectional Analysis.” The Lancet Public Health 1 (2): e76–83.
Peyré, Gabriel, Marco Cuturi, et al. 2019. “Computational Optimal Transport: With Applications to Data Science.” Foundations and Trends in Machine Learning 11 (5-6): 355–607.
Srivastava, Anuj, and Eric P Klassen. 2016. Functional and Shape Data Analysis. Vol. 1. Springer.