Mesoscopic mapping of the human structural connectome using high-performance global tractography

Overview & Synopsis

Mapping the human structural connectome is one of the major challenges of neuroimaging. Faced with the complexity of neuronal connections, imaging methods are constantly evolving to reveal ever-finer details. From Klingler dissections to diffusion MRI and advanced microscopy methods, these complementary modalities enhance our understanding of white matter architecture at multiple scales. Tractography has been introduced as a computational method for the virtual reconstruction of axonal fibres from diffusion-weighted MRI data. Among these methods, the global tractography method is based on the use of spin glasses, small portions of axonal fibres, whose positions, orientations and connections are variables jointly optimised by means of a cost function. Unlike conventional methods, which estimate these variables individually for each fibre, this approach identifies a global optimum tending towards the most plausible fibre configurations given the axonal directions derived from diffusion MRI. The underlying Metropolis-Hastings optimisation process remains computationally very expensive, preventing any large-scale use. To this end, ExaTract was developed as part of this thesis to considerably speed up computations, taking advantage of the emergence of HPC architectures. ExaTract can be applied both to conventional in vivo diffusion MRI datasets and to very high-resolution datasets from high-field MRI or 3D-PLI microscopy. Whereas wide phenotyping approaches aim to acquire data on large cohorts of subjects to extract patterns, deep phenotyping focuses on precise and exhaustive mapping of a small number of specimens. The Chenonceau project took this approach: by pushing the boundaries of high-field MRI, a post-mortem human brain was acquired over two campaigns spread over nearly 2 years and carried out jointly on NeuroSpin's 7 and 11.7 Tesla preclinical MRIs, achieving up to 100 µm resolutions. This unique in-the-world multi-modal MRI dataset gathers 48 fields of view covering the whole brain and provides cyto- and myelo-architectonic features and structural connectivity at the mesoscopic scale. The contribution of this thesis to the Chenonceau project concerns data processing and the reconstruction of this massive dataset at a whole-brain scale, allowing it to be shared with the scientific community. In a third aspect, ExaTract was applied to the Chenonceau dataset to establish the first atlas of human brain connectivity on a mesoscopic scale. Using unsupervised clustering methods, the built atlas includes not only deep white matter bundles but also short fibres, which were observed on a scale never met before. Although poorly studied in the literature, these fibres were found in very large numbers in the Chenonceau brain.