CROSS-TRACTS

Revealing the brain’s white matter crossing fibers’ topology: toward a new generation of tractography algorithms integrating the ground truth neuroanatomy.

In the innovative connectomics field, only diffusion magnetic resonance imaging (dMRI) tractography allows the building of a complete structural connectome non-invasively. However, dMRI tractography results are still controversial compared to the ground truth white matter (WM) anatomy. Diffusion MRI tractography algorithms aim at inferring WM fiber direction information based uniquely on a water diffusion displacement profile and thus fail to reconstruct complex WM crossings pathways reliably. The current CROSS-TRACTS project aims to reveal the topology of the WM crossing fibers in the mouse brain at an unprecedented accuracy, thanks to advanced multimodal tractography approaches. We first apply a gold-standard methodological approach for unraveling the WM’s anatomical features by combining viral tract labeling, whole-brain clearing, and light-sheet microscopy imaging (LSI). Second, in addition to advanced dMRI tractography, we use rs-fMRI to investigate WM fibers’ fanning, bending, or 3-way asymmetric crossings at the macroscopic scale. CROSS-TRACTS will thus produce tractograms across different modalities (dMRI, rs-fMRI, LSI) and at multiple scales (macroscopic, mesoscopic). We will tackle the compound challenge of integrating tractograms across modalities by applying a new deep neural network-based methodology that combines the dMRI-, rs-fMRI- and LSI-based tractograms within a correspondence autoencoding architecture. To do so, we allie scientific partners with very complementary expertise to resolve different technical issues.

CROSS-TRACTS has received the support of the ANR (ANR-22-CE45-0004) with a funding of 630 000 € for 48 months up to September 2026.

  • We will thus provide the neuroscience community with groundbreaking anatomical knowledge of the WM crossings’ topology.
  • We will provide the “neural-tracing/LSI” community with an unprecedented LSI-based tractography tool to track their cleared fluorescent samples.
  • We will give the fMRI community a dedicated fMRI-based asymmetrical tractography algorithm.
  • We will provide the dMRI community with a new generation of tractography algorithms that actually considers the ground truth of WM anatomy through a deep neural network autoencoder framework.

CROSS-TRACTS has a strong potential for back-translation to human studies. Our preclinical approach is crucial in developing tools for integrating prior knowledge in dMRI tractography, and this can be extensively applied to clinical studies and large multimodal neuroimaging databases.

 

Objectives

In the last decade, significant efforts have been dedicated to building and studying comprehensive maps of complex wiring diagrams of the brain, aka the connectome. In this innovative connectomics field, diffusion-weighted magnetic resonance imaging (dMRI) and resting-state functional MRI (rs-fMRI) are the leading non-invasive imaging technologies. In particular, dMRI tractography allows non-invasive delineation of white matter (WM) fiber pathways, previously not accessible out of post-mortem dissection studies, opening up vast possibilities for basic, preclinical and clinical research. However, the major problem with the existing white matter fiber tracking methodology remains its accuracy relative to ground truth WM anatomy (Maier-Hein et al. 2017; Thomas et al. 2014).

Our previous studies challenged the current state-of-the-art dMRI tractography and recommended focusing on tighter integration of anatomical priors in the tractography algorithms (Maier-Hein et al. 2017; Schilling et al. 2020; Schilling et al. 2021a; Schiavi et al. 2020). Currently, most dMRI tractography algorithms aim at inferring WM fiber direction information based solely on a water diffusion displacement profile and fail to reliably reconstruct complex crossings and bottlenecks of WM pathways (Schilling et al. 2021b). There is, therefore, an urgent need for methodological innovation in tractography to advance our knowledge of the human white matter anatomy and to build anatomically correct connectomes. CROSS-TRACTS addresses this challenge by studying for the first time the topology of the brain’s white matter crossing fibers at an unprecedented degree of accuracy thanks to advanced multimodal tractography approaches. To the best of our knowledge, no previous work has clearly described how, from an anatomical perspective, the WM fibers cross within bundles. For example, the fibers constituting the pyramidal (PYT), arcuate (AF), and callosal (CC) pathways intersect in human WM within the centrum semiovale. The thick antero-posterior AF is crossed by fanning dorso-ventral PYT and medio-lateral CC fibers. How are these crossings organized? Is it random? Or is there a structured organization of these crossings, for example, in interlacing thin layers of fibers? We previously performed a detailed dissection of this region (De Benedictis et al. 2016) and observed the existence of such layers. But only an approach that enables the visualization of this structure at the mesoscopic scale will provide a definitive answer to these questions, which requests an animal model. Thus, CROSS-TRACTS aims at describing the organization of crossings between WM fibers in the mouse brain.

To achieve this original and novel goal, we will first apply a gold-standard methodological approach for unraveling the WM’s anatomical features by combining viral tract labeling, whole-brain clearing, and light-sheet microscopy imaging (LSI). LSI uses a light-sheet laser technique to image an anatomically intact transparent mouse brain at axonal resolution. This approach will be especially beneficial for visualizing, at the mesoscopic scale, the complex organization of crossing WM fibers across the entire brain through an innovative LSI-based tractography approach. For example, if PYT is labeled in green and CC in red, we will not only be able to visualize both of them but also resolve their 3D crossings at the mesoscopic scale.

Second, in addition to advanced WM fiber reconstruction in dMRI tractography, CROSS-TRACTS will investigate the complex WM organization at the macroscopic scale by using rs-fMRI. We will use the asymmetric temporal correlation distributions of rs-fMRI signal to untangle the axonal configuration containing a fanning, bending, or a 3-way asymmetric crossing. It has been shown that neighborhood temporal correlations of rs-fMRI signals display distinct anisotropy within WM regions (Ding et al. 2013). Such asymmetric directional variations of rs-fMRI temporal correlations are highly consistent with the trajectory of expected underlying WM pathways (Schilling et al. 2019).

CROSS-TRACTS will thus produce tractograms across different modalities (dMRI, rs-fMRI, LSI) and at multiple scales (macroscopic, mesoscopic). Many modality-specific tools exist to process imaging data, but integrating across modalities presents a compound challenge (Goubran et al. 2019). We will tackle it by introducing a mathematical formalism that will combine the dMRI-, rs-fMRI- and LSI-based tractograms within a correspondence autoencoding architecture (cf. graphical summary below). The autoencoders (AEs) allow low dimensional representations of a dataset, i.e., a tractogram made of WM fibers. Once the different encoding is obtained across modalities, we will take advantage of a coupled-AEs approach. It involves embedding each tractography dataset (dMRI, rs-fMRI, and LSI) into a shared space representing the latent state of the different multimodal tractograms, such as the distributions of each dataset mapped into the latent space is aligned. Combining the encoder and decoder modules of different autoencoders enables translation between different tractography data modalities. As shown below, the ground truth WM crossing fibers obtained with LSI-based tractography will allow building dMRI-based tractograms using crossings’ topology priors. CROSS-TRACTS partners have already demonstrated that AEs can learn how to filter WM fibers to obtain more reliable tractograms even with a limited amount of data in the training set (Legarreta et al. 2021), which is an issue to consider with our preclinical approach.

CROSS-TRACTS will use macroscopic and mesoscopic imaging techniques that each project partner well masters. The challenge lies in all data acquisition will be applied (WP1), processed, and combined (WP2) in the same mouse brains. The whole WP1 will be dedicated to the different data acquisitions. It aims at feeding the multimodal tractography approaches that the WP2 will address.

WP1: Acquisition of the macroscopic and mesoscopic imaging data of the crossing WM fibers.

  1. In vivo dMRI and rs-fMRI acquisition in each animal will be performed at the Plateforme d’Imagerie Biomédicale facility (PIBio, UMS3767, www.pibio-bordeaux.cnrs.fr) on a 7-Tesla small animal Bruker magnet equipped with an intense gradient system and a cryoprobe coil. Whole-brain rs-fMRI data will be acquired with a spatial sampling of 200 x 200 x 400 µm3, a temporal sampling of 2 sec for a total duration of 8 min. dMRI data will be acquired with a single-shot spin-echo EPI sequence (isotropic spatial resolution of 100 µm in approximately 30 minutes with one b-value and 60 directions. A self-gating module will be implemented into the sequence to limit the impact of breathing and allow correcting motion corrupted data, a posteriori (Ribot et al. 2015). 3D high-resolution T1- and T2-weighted anatomical acquisitions will also be performed notably for the dMRI and rs-fMRI data’s co-registration. A high-resolution post-mortem dMRI acquisition (803 µm3) will also be performed subsequently after the sacrifice of the animal.
  2. Viral tract labeling, whole-brain clearing, and light-sheet microscopy. We will use a multi-labeling approach to simultaneously visualize different axonal projections and their crossings within the WM. Two AAV viruses expressing one specific fluorescent marker (dtomato, eGFP, mKate, or YFP) under the control of the synapsin or CBA promotor (serotype will depend on brain region) will be stereotaxically injected into different brain regions of adult C57BL/6 mice (Haberl et al. 2015). Following four weeks of viral expression, mice will be perfused for whole-brain clearing using the iDISCO method. Previously used by the Frick team and the Bordeaux Imaging Center to clear entire mouse brains, iDISCO preserves the markers’ endogenous fluorescence at least for several weeks. Cleared-whole brains will then be imaged using LSI at a spatial resolution that enables visualization of WM fibers.

Preliminary results obtained by the CROSS-TRACTS Partners. A. Diffusion MRI-based tractography of an high-resolution (73 µm isotropic) postmortem mouse brain acquired in the Plateforme d’Imagerie Biomédicale (UAR3767, www.pibio-bordeaux.cnrs.fr) on a 7-Tesla small animal Bruker magnet with a cryoprobe coil; The bottom part of A shows a crossing region where some association fibers (in green) cross the callosal fibers (in red). B. Whole mouse brain obtained in the Bordeaux Imaging Center facility (BIC, C. Poujol and J. Teillon, Centre Broca Nouvelle-Aquitaine, www.bic.u-bordeaux.fr/) with LSI after viral injection of the right dentate gyrus in the hippocampus and whole-brain clearing (courtesy of S. Desforges, IINS, UMR5297). It enables the visualization of actual fibers connecting the two hippocampi. C. Dual-labeling of reciprocal cortico-thalamic connections.Two recombinant rabies virus variants expressing either mCherry (anterogradely transported) or eGFP (retrogradely transported) were injected into the thalamus. The anterogradely labeled projections (red) ascend from neurons of the injection site, the VPm region, and arrive in the barrel cortex (S1 BC). The retrogradely labeled cells are concentrated in layers 5 and 6 (green) of S1 BC, layers known to provide feedback to the thalamus.

WP2: Multimodal tractography of the crossing WM fibers

  1. dMRI-based tractography: Both in-vivo and ex-vivo whole-brain dMRI tractograms will be first built with high angular resolution diffusion imaging (HARDI) tractography where the tracking directions are determined from the fiber orientation distributions (fb-ODs, (Descoteaux et al. 2009)). Projection and commissural WM pathways will be extracted with regions of interest positioning guided by existing anatomical mouse brain atlas. Cortical terminations of these bundles will determine the injection sites for the viral tract labeling. Therefore, we will apply for the first time in the mouse brain our new deep learning framework in tractography, FINTA (Legarreta et al. 2021). Working with simple architecture, FINTA can robustly learn the structure of streamlines in tractography, then apply such a framework to discriminate between anatomically plausible and implausible streamlines.
  2. rs-fMRI-based tractography: Anisotropic correlations of rs-fMRI signals between WM voxels will be modeled as a functional orientation distributions map (rs-ODs). RS-ODs map will provide asymmetric bending and fanning distribution to be tested as directional priors used in the dMRI tractography and compared to the shape of streamlines obtained with LSI-based tractography within the coupled-AEs formalism (WP2.4).
  3. LSI-based tractography: Reconstructing axonal projection to streamlines from clarified mouse brain tissue calls for hitherto underdeveloped tractography techniques that CROSS-TRACTS will highlight. LSI-based tractograms will be reconstructed from 3D images of cleared whole brains by recovering local fb-ODs from image intensity gradients analysis. We recently developed a similar method to target fiber crossings based on two retrograde viral injection site locations (Lefebvre et al. 2021).
  4. Integration of tractography data across multi-scale modalities: CROSS-TRACTS will develop a new deep neural network-based methodology referred to as coupled-AEs for the first time in the tractography domain. Coupled-AEs consist of multiple AE networks, each comprising encoder and decoder subnetworks that are nonlinear transformations projecting input data into a low-dimensional representation, the integrated latent space. Here, dMRI-, rs-fMRI-, and LSI-tractograms will constitute three AE networks. In learning the transformations, the goal will be to simultaneously maximize reconstruction accuracy for each data modality and similarity across representations for the different modalities. In other words, coupled AEs can facilitate streamlines/fibers correspondence across modalities (cf. graphical summary on page 2).

 

Partnership

CROSS-TRACTS is carried out in close collaboration between different imaging teams gathered in the Bordeaux Neurocampus: the Groupe d’Imagerie Neurofonctionnelle (GIN-IMN-UMR5293), the Centre de Résonance Magnétique des Systèmes Biologiques (RMSB-UMR5536), the Cortical Plasticity (Frick) team (Neurocentre Magendie) and the Bordeaux Imaging Center (BIC). GIN-IMN researchers involved in CROSS-TRACTS are experts in dMRI-tractography and rs-fMRI. RMSB researchers are specialized in the development of innovative strategies for ultra-fast preclinical imaging in whole-body mice. The Frick team has many years of expertise in using viral tracers to explore neuronal circuits at the nano-, micro-, and mesoscale levels, various clearing methods. The BIC is a core facility that provides a unique set of high-end equipment in photonic and electronic imaging, mainly in animal and plant biology.

  • Laurent Petit, coordinator (CNRS Researcher, UMR5293-CNRS/CEA/Univ.Bordeaux) has published more than 100 peer-reviewed articles, including 35 dedicated to the dMRI-tractography since the last three years, with 50% including ex vivo microscopic dissection data for validation of dMRI tractography. He masters brain anatomy and developed innovative strategies in applying multimodal tractography to reveal WM anatomy. Role in the project: Project management, supervision of MRI data acquisition and tractography.
  • Marc Joliot (CEA Research Director, UMR5293-CNRS/CEA/Univ.Bordeaux) has authored 125 peer-reviewed articles. He is mastering and developing methods for rs-fMRI processing on human data, also transferred to the preclinical domain in collaboration with the PIBio (UMS3767). PI of MIMACORE project (IMAG’IN, CNRS 2017) studying neuro-metabolic micro-macroscopic anatomo-functional connectivity in rodents. PI of the LabCom GINESISLAB (2015-2021) developing deep-learning processing of medical images (Boutinaud et al. 2021). Role in the project: Supervision of rs-fMRI tractography and multimodal tractography integration
  • Sylvain Miraux (CNRS Research Director, UMR5536-CNRS/Univ.Bordeaux), leading the RMSB, has published more than 60 peer-reviewed articles in MRI methodology and its applications and has experience in MR sequence development at a high magnetic field on preclinical small animal models. He has experience in project management in Technology for Health (ANR TecSAN) and developed collaborations with industrial partners (Bruker, Siemens). Role in the project: Supervision of dMRI and rs-fMRI acquisition.
  • Andreas Frick (INSERM Research Director, INSERM U1215, Univ.Bordeaux), leading the Cortical Plasticity Group at the Neurocentre Magendie, is an expert in analyzing the functional and structural connectivity of neocortical circuits and their reorganization in brain disorders and memory formation. Role in the project: Supervision of viral tracing, brain clearing, and LSI acquisition and analysis with the BIC.
  • Other people involved in the project: GIN-IMN: Maxime Descoteaux (SCIL, Université de Sherbrooke, Canada) experience in development of tractography algorithm (via pre-existing collaboration); RMSB: Aurélien Trotier (CNRS engineer), experience in preclinical imaging and MR sequence development; William Lefrançois (Assist. Prof. Univ. Bordeaux) specialized in fast acquisition strategies; Frick team: Melanie Ginger (IR, INSERM), expertise in production and in vivo exploitation of viral vectors; Christel Poujol and Jérémie Teillon (BIC, Bordeaux) experts in LSI acquisition and analysis.

References 

In color the articles from the CROSS-TRACTS partners

Boutinaud P, Tsuchida A, Laurent A, Adonias F, Hanifehlou Z, Nozais V, Verrecchia V, Lampe L, Zhang J, Zhu Y-C, Tzourio C, Mazoyer B, Joliot M (2021) 3D segmentation of perivascular spaces on T1-weighted 3 Tesla MR images with a convolutional autoencoder and a U-shaped neural network. Frontiers in Neuroinformatics 15 (29). doi:https://doi.org/10.3389/fninf.2021.641600

De Benedictis A, Petit L, Descoteaux M, Marras CE, Barbareschi M, Corsini F, Dallabona M, Chioffi F, Sarubbo S (2016) New insights in the homotopic and heterotopic connectivity of the frontal part of the human corpus callosum revealed by microdissection and diffusion tractography. Hum Brain Mapp 37 (12):4718-4735. doi:http://dx.doi.org/10.1002/hbm.23339

Descoteaux M, Deriche R, Knosche TR, Anwander A (2009) Deterministic and probabilistic tractography based on complex fibre orientation distributions. IEEE transactions on medical imaging 28 (2):269-286. doi:http://dx.doi.org/10.1109/TMI.2008.2004424

Ding Z, Newton AT, Xu R, Anderson AW, Morgan VL, Gore JC (2013) Spatio-Temporal Correlation Tensors Reveal Functional Structure in Human Brain. PLoS ONE 8 (12):e82107. doi:http://dx.doi.org/10.1371/journal.pone.0082107

Goubran M, Leuze C, Hsueh B, Aswendt M, Ye L, Tian Q, Cheng MY, Crow A, Steinberg GK, McNab JA, Deisseroth K, Zeineh M (2019) Multimodal image registration and connectivity analysis for integration of connectomic data from microscopy to MRI. Nature Communications 10 (1):5504. doi:http://dx.doi.org/10.1038/s41467-019-13374-0

Haberl MG, Zerbi V, Veltien A, Ginger M, Heerschap A, Frick A (2015) Structural-functional connectivity deficits of neocortical circuits in the Fmr1 -/y mouse model of autism. Sci Adv 1 (10):e1500775. doi:http://dx.doi.org/10.1126/sciadv.1500775

Lefebvre J, Delafontaine-Martel P, Lemieux P, Descoteaux M, Petit L, Lesage F (2021) Localization and imaging of white matter fiber crossings in whole mouse brains using diffusion MRI and serial blockface OCT, vol 11629. SPIE BiOS. SPIE. doi:https://doi.org/10.1117/12.2577648

Legarreta JH, Petit L, Rheault F, Theaud G, Lemaire C, Descoteaux M, Jodoin P-M (2021) Filtering in tractography using autoencoders (FINTA). Medical Image Analysis 72:102126. doi:https://doi.org/10.1016/j.media.2021.102126

Maier-Hein KH et al. (2017) The challenge of mapping the human connectome based on diffusion tractography. Nat Commun 8 (1):1349. doi:http://dx.doi.org/10.1038/s41467-017-01285-x

Ribot EJ, Duriez TJ, Trotier AJ, Thiaudiere E, Franconi J-M, Miraux S (2015) Self-gated bSSFP sequences to detect iron-labeled cancer cells and/or metastases in vivo in mouse liver at 7 Tesla. Journal of Magnetic Resonance Imaging 41 (5):1413-1421. doi:http://dx.doi.org/doi:10.1002/jmri.24688

Schiavi S, Ocampo-Pineda M, Barakovic M, Petit L, Descoteaux M, Thiran J-P, Daducci A (2020) A new method for accurate in vivo mapping of human brain connections using microstructural and anatomical information. Science Advances 6 (31):eaba8245. doi:http://dx.doi.org/10.1126/sciadv.aba8245

Schilling KG, Gao Y, Li M, Wu TL, Blaber J, Landman BA, Anderson AW, Ding Z, Gore JC (2019) Functional tractography of white matter by high angular resolution functional-correlation imaging (HARFI). Magn Reson Med 81 (3):2011-2024. doi:http://dx.doi.org/10.1002/mrm.27512

Schilling KG, Petit L, Rheault F, Remedios S, Pierpaoli C, Anderson AW, Landman BA, Descoteaux M (2020) Brain connections derived from diffusion MRI tractography can be highly anatomically accurate-if we know where white matter pathways start, where they end, and where they do not go. Brain Struct Funct 225 (8):2387-2402. doi:http://dx.doi.org/10.1007/s00429-020-02129-z

Schilling KG et al. (2021a) Tractography dissection variability: What happens when 42 groups dissect 14 white matter bundles on the same dataset? Neuroimage 243:118502. doi:https://doi.org/10.1016/j.neuroimage.2021.118502

Schilling KG, Tax CMW, Rheault F, Landman B, Anderson A, Descoteaux M, Petit L (2021b) Prevalence of white matter pathways coming into a single diffusion MRI voxel orientation: the bottleneck issue in tractography. Hum Brain Mapp in press. doi:https://doi.org/10.1101/2021.06.22.449454

Thomas C, Ye FQ, Irfanoglu MO, Modi P, Saleem KS, Leopold DA, Pierpaoli C (2014) Anatomical accuracy of brain connections derived from diffusion MRI tractography is inherently limited. Proc Natl Acad Sci U S A 111 (46):16574-16579. doi:http://dx.doi.org/10.1073/pnas.1405672111