Contents
The Diffusion simulation for tissuE miCrostructure and bRain connectivitY with oPtimized acquisiTions (DECRYPT) is an associate team between Empenn (Inria Rennes Bretagne Atlantique) and LTS5 (École Polytechnique Fédérale de Lausanne — EPFL) started in 2024.
Team members
Élise Bannier (MR Physicist, Empenn/Neurinfo, CHU Rennes)
Constance Bocquillon (PhD student, Univ Rennes)
Erick Jorge Canales-Rodríguez (SNF Ambizione Fellow, LTS5, EPFL)
Emmanuel Caruyer (Researcher, Empenn, CNRS)
Julie Coloigner (Researcher, Empenn, CNRS)
Isablle Corouge (Research engineer, Empenn, Univ Rennes)
Sébastien Dam (PhD student, Empenn, Inria)
Elda Fischi Gomez (Scientist, LTS5, EPFL)
Pierre Maurel (Professor, Empenn, Univ Rennes)
Marco Pizzolato (assistant professor, DTU Copenhagen/EPFL)
Marie Poirier (PhD student, Empenn/LMJL, Univ Rennes)
Benjamin Prigent (PhD student, Empenn, Inria)
Adèle Savalle (PhD student, Empenn, Inria)
Jonathan Rafael Patiño (Postdoc, LTS5, EPFL/CHUV)
Ekin Taskin (PhD student, LTS5, EPFL)
Jean-Philippe Thiran (Professor, LTS5, EPFL)
Juan Luis Villarreal Haro (PhD student, LTS5, EPFL)
Grégoire Ville (PhD student, Empenn, Inria)
Context
The reconstruction of microscopic-scale information using magnetic resonance and its application to biological tissue in vivo with magnetic resonance imaging (MRI) has boosted our understanding of the organization of organs, in health and pathology (Alexander et al., 2019). In particular, neuroimaging with diffusion MRI has unveiled unprecedented details on the brain architecture with white matter tractography and the analysis of the brain connectome. In neuro-degenerative diseases, microstructural alterations usually occur at a relatively early stage, their detection could provide unique insight for the diagnosis, prognosis and monitoring of a number of pathologies, including but not limited to multiple sclerosis, Alzheimer’s disease, stroke or patients in a coma.
A number of challenges have been highlighted in this quest for a microstructure-informed connectome reconstruction (Maier-Hein et al., 2019). In particular, current methods lack specificity and sensitivity to parameters of interest. These parameters are either local descriptors of the microstructure (cellular density, shape/caliber parameters) or macroscopic descriptors of the brain connectivity (detection and quantification of the “strength” of the connectivity between two interconnected brain regions). By developing numerical substrate, Monte-Carlo simulation methods, inverse problems solving with fingerprinting and machine learning, and the design of data acquisition methods tailored for specific tasks in microstructure characterization, we will contribute to this developing field. We will also develop methods for the integrated statistical analysis of the microstructure-informed brain connectome, building on recent development of graph-based signal processing and statistics on functional data.
Objectives
In the context of the DECRYPT associate team, we will contribute to the development of diffusion simulation for the analysis of tissue microstructure and brain connectivity with optimized acquisitions with the following work packages.
WP1: substrates design and Monte-Carlo simulations
Using an approach inspired by fingerprinting, we will generate a collection of substrates that follow a range of microstructure parameters, and simulate the diffusion signal that would be measured in each of these substrates in isolation. This will define a dictionary of signal fingerprints that can be combined linearly to explain the signal in a complex configuration, with partial volume occupied by different tissue microstructure. A key element in this work package will be to control the random effects of substrates generation and evaluate the critical size of the substrates so that the predicted signal is reproducible across substrates generated from repeated parameters. A range of microstructure and diffusion/relaxation parameters will be incorporated in the model, so that the microstructural effects related with pathology evolution (e.g. axonal loss, myelin sheath damage, membrane permeability changes, cellular density, …) can be simulated.
WP2: acquisition design for optimized sensitivity and specificity
In order to identify specific substrate configurations from the diffusion-weighted MRI signal, it is important that the acquisition parameters are designed and tailored to be sensitive and specific to the microstructure parameters of interest. Optimizing on diffusion-encoding parameters is a complex mathematical problem, due to the large dimension of the space of acquisition parameters. We will build upon recent methods developed by the partners to design specific diffusion-encoding sequences, using b-tensor encoding and free gradient waveforms, that enable the accurate estimation of microstructure parameters.
WP3: connectivity analysis via complex network-based statistics
When projected along the trajectories of white matter pathways, microstructure measures define a network in which every edge is valued by a set of functional data, representing the evolution of parameters along the path of the curve. This defines a graph, where each edge is associated with a rich information, which can be represented as a multi-valued function along the path of the fiber. In order to reach the full potential of this complex information, we will develop specific statistical analysis methods, building upon recent results in brain network analysis and graph signal processing.
2019-2023: The MMINCARAV associate team
Prior to DECRYPT, the Multimodal Microstructure-Informed Connectivity: Acquisition, Reconstruction, Analysis and Validation (MMINCARAV) project was an associate team between Empenn (Inria Rennes Bretagne Atlantique) and LTS5 (École Polytechnique Fédérale de Lausanne — EPFL). The main objective of this associate team was to develop novel methods for the quantification of microstructure properties in brain tissues using diffusion MRI and relaxometry, in relation with the macrostructure organization of the brain connectome. The project was organized into 4 work packages:
- Acquisition design in microstructure-enabled diffusion MRI using compressive sensing
- Combining relaxometry and diffusion
- Analyzing the microstructure-informed brain connectome
- Validating microstructure and connectivity reconstruction using simulations and reproducibility analysis
Context and motivations of the MMINCARAV project
The ability to measure microscopic scale phenomena using magnetic resonance imaging (MRI) has triggered a revolution in neuroimaging, and brought hope for a better understanding of the brain’s structure and function. In particular, the recent developments of diffusion MRI and relaxometry have shown the tight link between tissue micro-organization and the measured signal. Due to their relatively large volume (~1mm³), there is a mix of tissue properties in every voxel in the image, and several bio-physical models have been proposed to encompass this heterogeneity. Besides, using this local information on microstructure, it is not only possible to better estimate the trajectory of major fiber pathways, but also to have a more relevant, quantitative description of the brain connectivity.
There is a large number of open problems to this end; first, recovering tissue properties from the measured signal requires solving an inverse problem, which is in general ill-posed due to the relatively small number of signal samples we can acquire during an in vivo scan (10 minutes up to one hour). A proper design of the acquisition sequence is of utmost importance in this respect, to make sure we take the best advantage of every single measurement. In addition, relaxometry and diffusion MRI provide complementary information; despite the growing interest, realistic models accounting for both relaxometry and diffusion remain to be developed to better characterize the brain microstructure. Progress in the microstructure modelling paves the way for quantitative connectivity measures, directly related to the actual number of axons connecting two regions. Computing and analyzing the microstructure-informed brain connectome, using methods of graph theory, opens new challenges since the statistical properties of this connectome are very different from current connectivity matrices, for which weights are computed counting streamlines of a tractogram. Last, recent contests organized in international conferences have shown the limits of state-of-the-art techniques in controlling the number of false positives when estimating the connectome, and there is a pressing demand in validating the reconstruction from the micro- to the macroscopic scale.
Objectives of the MMINCARAV associate team
We propose to address these challenges with the following work packages (WP).
WP1: Acquisition design
(key investigators: E. Caruyer, M. Pizzolato, R. Truffet)
In diffusion MRI, the use of generalized diffusion-weighting gradient waveforms was shown to provide higher sensitivity to microscopic tissue properties. While in practice most sampling schemes still use rectangular pulsed gradients due to their simplicity, we will exploit the larger number of degrees of freedom to design a family of gradient waveforms adapted to microstructure reconstruction. We will use the theory of compressive sensing to select gradients which are best adapted to reconstruction using sparse priors.
WP2: Combining relaxometry and diffusion
(key investigators: C. Barillot, O. Commowick, M. Pizzolato, T. Yu)
Diffusion MRI allows disentangling the signal contributions coming from the intracellular and interstitial water molecular dynamics, but is generally blind to water trapped in myelin sheets. Conversely, T2-relaxometry is blind to the separation of intracellular and interstitial water but it allows disentangling their joint contribution from that of myelin. We will exploit the complementarity of these two modalities by proposing a unified model of extended microstructure that includes these three major compartments. The pertinence of this model will be evaluated in healthy controls and patients.
WP3: Analyzing the microstructure-informed brain connectome
(key investigators: J. Coloigner, C. Barillot, G. Girard, J.-Ph. Thiran)
Graph theory provides powerful methods for evaluating structural brain networks.. This technique models the brain as a complex network where nodes are associated with regions of interest and an edge strength represents the degree of structural connection between a pair of regions. Conventional approaches use connectivity measures derived from the number of streamlines estimated by tractography algorithms. However, false positive connections arise frequently and can directly bias the inferred topology measures. Microstructure features, such as the mean axon diameter, have been shown to reduce false positive connections and biases in the estimated structural connectivity. We will develop new network analysis methods based on the measure of the microstructure-informed connectivity, and evaluate how this naturally reduces the biases.
WP4: Validating microstructure and connectivity reconstruction
(key investigators: E. Caruyer, G. Girard, J. Rafael Patiño, J.-Ph. Thiran)
The diffusion MRI community is in a pressing need for validation methods for sampling efficiency, microstructure model accuracy, and novel connectivity measures. We will build upon the respective expertise of both LTS5 and VisAGeS in simulating diffusion MRI at the mesoscopic scale and the macroscopic scale to create a macroscopic simulator using locally realistic microstructure modelling. We will also implement the designed acquisition sequences on a small animal MRI scanner and a human MRI scanner, and acquire a test-retest dataset to perform a reproducibility study. This will set the basis for the organization of a reconstruction challenge on connectivity using realistic data simulations.
