2016 School on Brain Connectomics

Diffusion MRI (dMRI) is the unique Magnetic Resonance Imaging modality able to quantify in vivo and non invasively the average random thermal movement (diffusion) of water molecules in biological tissues such as brain white matter. Using the water diffusion as a probe, dMRI makes it possible to reconstruct white matter fiber pathways and segment major fiber bundles that reflect the structures in the brain which are not visible to other non-invasive imaging modalities. This modern imaging modality, of great interest to neuroscientists and clinicians, has opened a number of challenging problems. In this talk, the important problems of efficiently acquiring and processing complex dMRI data will be introduced and recently developed solutions and advances will be presented. Applications to computational brain imaging will also be presented and discussed with a particular emphasize on the importance of the Riemannian geometry in the estimation, regularization and segmentation of diffusion images as well as the tracking, the reconstruction and the clustering of the bundles of white matter fibers. High Angular Resolution Diffusion Imaging (HARDI) models will also be presented to go beyond the classical Diffusion Tensor Model, well known to be inadequate in crossing fiber regions. These new algorithms open the possibility of inferring and recovering a more detailed geometric description of the anatomical connectivity between brain areas. The presentation of some open problems currently being investigated by the dMRI community will conclude the talk.

In this introductory lecture I will review the basic principles of structural brain connectivity analysis by diffusion MRI, including the physical principles of diffusion, the most common MR acquisition schemes, the main principles of tractography and network analysis that lead to the concept of connectomics. Then I will review some of the applications of connectomics in clinical research. Finally, I will underline the main remaining technical challenges in brain connectivity analysis by diffusion MRI. Some of them will be addressed in detail in other lectures during this summer school, and I will briefly introduce them and place them in their context.

Diffusion MRI offers a way to study the microstructure of living tissue in vivo. Tissue properties such as cell sizes, shapes and orientations influence the water diffusion. These tissue properties can thus be inferred from the MRI data, by encoding information about the diffusion using time-varying magnetic field gradients. Diffusion MRI was traditionally performed with just one set of gradient timing parameters, which limit the information that can be retrieved from the imaging data. Recently, novel approaches have demonstrated that extended information on the tissue microstructure can be retrieved by using multiple sets of gradient timing parameters. In this talk, I will cover the basics of relating diffusion in simple geometries to the observed signal, which serve as the foundation for so-called compartment models. I will describe experimental requirements and limitations for retrieving microstructure information from diffusion MRI. Finally, I will also cover recent developments of using so-called b-tensor encoding to extend the scope of microstructure information retrievable from diffusion MRI.

The study of brain connections has a long history dating back to postmortem blunt dissections performed by pioneer anatomists of the 18th and 19th century. This method, has allowed researchers to identify the trajectories of the major white matter bundles of the human brain. The development of animal tracing studies in the 20th century has further improved our knowledge of connections, adding more detailed descriptions of brain connectivity but these studies must be performed in non-human animals and their findings in other species remain to be confirmed.

In 1994, a real breakthrough came with the development of diffusion tensor imaging (DTI) to study the organization of white matter in the living human brain. With this imaging method and the development of tractography it was possible, for the first time, to map in vivo and noninvasively the organization of white matter fibers. Rapidly, this technique has become the most important tool for investigating the connectional anatomy of the normal and pathological human brain. By extracting quantitative diffusion indices along the dissected tract it has allowed the characterization of microstructural properties of tissue in the normal and pathological brain and has provided a way to extract quantitative measurements for group comparisons or individual case studies.

Over the years, new diffusion models have been introduced to overcome limitations of DTI and new tractography algorithms have been proposed to better reconstruct three dimensional trajectories of the major white matter pathways. Most of the current tractography algorithms are based on the same tracking strategies originally introduced by the first tractography approaches. However, when tractography is performed using multifibre approaches, new strategies are also necessary because of the increased risk of false-positive reconstructions. Different new approaches have been recently proposed to guide the propagation of the tractography algorithm across white matter regions and try to discriminate between complex tract configurations using tract specific and microstructure indices.

Despite these improvements, low spatial resolution of diffusion MR data and the lack of histological validation remain two of the major limitations of in-vivo diffusion tractography. Little is still known about the true correspondence between tractography reconstructions, diffusion indices and the real underlying tissue organization. While for some major tracts the details of these "virtual reconstructions" can match closely those derived from human post-mortem blunt dissections or histology, for other smaller or less common tracts, this matching is often incomplete or can produce a large number of false connections. Hence, validation of tractography reconstructions with complementary methods remain an essential step before new anatomical models can be put forward or applied to clinical populations.

Investigating the entire connectivity of the human brain is today one of the most challenging tasks in neuroscience. Developing and applying novel techniques to visualize and study large brain networks is essential to achieve a complete reconstruction of the human brain connectome. Whole brain tractography approaches combined with network analysis tools are currently in development and may offer in the future a new way to investigate connectivity in the healthy and pathological brain. However, today connectomics results strongly rely on the anatomical accuracy of tractography methods and their ability to describe correctly white matter trajectories between distant cortical regions. Inevitably, the limitations of current tractography algorithms can be an important bias for the final results. Improving tractography methods further is essential and probably the main major goal that need to be achieved for tractography to provide a solid foundation for the mapping the human connectome.

In conclusion, the ability the reproduce three-dimensional trajectories of white matter connections in the living human brain is a unique feature of diffusion tractography. This makes Diffusion Imaging and Tractography one of the most exciting and challenging fields in neuroimaging. New methods are now emerging at a very fast pace and they can offer new opportunity to map connections in normal and pathological brain. However, validation of all these methods remain an essential step before they can broadly applied in clinical and research applications.

Fiber tractography is at the heart of the fast growing field of connectomics for its non‐invasive mapping of brain connections and is increasingly used in neuroscience and medicine. Even so, the lack of ground truth on long‐range connectivity of the human brain makes it harder to quantitatively evaluate results and determine which discoveries are reliable. Using a highly realistic simulated human brain dataset with known ground truth, we organized an open international tractography challenge, with 96 distinct submissions from 20 research groups. Using commonly defined measures, we assessed how different processing steps influenced the results. The results reveal substantial numbers of false‐positive bundles that are not part of the ground truth and yet consistently extracted by most algorithms. Our findings may inspire future algorithmic developments and may help to test them with the framework presented here. A key challenge will be to control for false positives, while identifying the full extent of existing fiber bundles.

Despite connectomics offers an exquisite tool to investigate non-invasively the architecture of the neuronal connections of the brain, the reconstructions recovered with existing algorithms are not really quantitative. The information recovered by diffusion MRI tractography is orders of magnitude coarser than the actual size of the axons and each of the recovered tracts represents an unspecified set of coherent real fibers. Consequently, the measures of structural connectivity between different brain regions estimated with existing tractography algorithms are only indirectly related to the actual properties of the underlying neuronal connections. In this talk I will go through the major steps forward in tractography putting a particular emphasis on the biological plausibility of the estimated tractograms. I will first review the state-of-the-art approaches for estimating the connectivity and then I will also present some very recent ideas that, in the near future, may potentially enable truly quantitative and biologically meaningful analyses of brain connectivity.

In this talk I will be presenting cortical parcellation schemes and their use to infer cortical brain connectivity. The aim is to provide a overview of existing parcellation schemes that can be used in the context of connectivity inference, and to understand the fundamental and practical aspects of building a cortical parcellation. I will first introduce the motivations and the general concept of cortical parcellation. I will then discuss what can be expected from a cortical parcellation in terms of properties (e.g. resolution or geometry) and general relations with function, connectivity, of microstructure. We will then move on to the different types of information that can be used to parcellate the cortex and the different classes of atlases they lead to. Existing parcellation techniques and associated softwares will then be reviewed. I will conclude with a brief discussion on the inverse approach: building parcellation from connectivity information.

I will present an overview of the interest in acquiring data of multiple modalities simultaneously, and illustrate the associated technical challenges and some of the disadvantages. Conceptual models of data fusion will be described and illustrated in a range of applications, in the diseased and healthy brain. Emphasis will be put on combined EEG and fMRI measurements for cognitive and disease characterisation studies.

Over the past years, brain mapping techniques have been enhanced - and largely inspired by approaches from signal processing and network theory - to overcome the shortcomings of voxel-wise brain mapping. First of all, brain function does not only rely on local specialization, but also on interactions between brain regions, a principle known as functional integration. New approaches that quantify such interactions have allowed to identify networks of brain regions that exhibit temporally coherent signals. Second, the brain is always active, functionally and metabolically. Therefore, spontaneous activity between task periods cannot be considered as constant, but carries meaningful information, even during complete "resting state"; i.e., the subject is asked to relax and let his mind wander. Data-driven analysis has been used to extract resting-state networks that show close resemblance to functional networks extracted from task-evoked data. Moreover, these intrinsic functional networks are altered by almost any brain disease. Together with minimal patient compliance (no task needed), these networks are promising candidates for non-invasive biomarkers for diagnosis and prognosis in a single patient, and for facilitating the translation of fMRI into the clinical realm. We will give an overview of multivariate and graph-based techniques for fMRI data processing, as well as how machine-learning approaches can be made of good use in this field. The quest for better understanding of brain dynamics has also triggered new ways to approach functional connectivity, which is typically measured as the correlation between timecourses over the whole run. Non-stationarity implies that FC should be fluctuating and that these fluctuations carry information about the underlying brain processes.

The characterization of brain dynamics and their relationship with the underlying anatomical wiring is central for our understanding of brain communication mechanisms. In this talk I will first discuss anatomical connectivity impairments in schizophrenia. Next, I will propose to integrate slow neural oscillations and anatomical-connectivity maps, as derived from functional and diffusion MRI, into a single multilayer graph framework to capture transient networks of spatio-temporal coherence. Within these networks, activity propagates in a shrinking-expanding mode through multi-step pathways that include selective ensembles of structural connections. The identified spatio-temporal networks could encode communication channels' selection and be relevant to the investigation of pathological conditions such as schizophrenia.

The brain is thought to consist of a network of interconnected, interacting components whose architecture is critical for the emergence of adaptive behaviors and cognition. Graph theoretical analyses provide a powerful framework to assess topology and organization of brain connectivity networks, and have contributed substantially to our understanding of the topological organization of brain connectivity, revealing a small-world, rich-club structure and the presence of hubs characterized by high centrality and connectivity. Moreover, a number of studies have investigated the modular structure of brain connectivity networks, highlighting cohesive clusters of nodes that are more densely connected among themselves than with the rest of the network. Topological modularity is thought to reflect functional and anatomical segregation, a feature that may confer robustness and adaptivity to brain networks. Moreover, the degree of clustering within functional connectivity graphs may provide a measure of the balance between segregation and integration underlying brain function. Here, I will present the principles of graph theoretical methods for the analysis of network modularity, and discuss their merits and limitations in the context of the application to brain connectivity networks. Moreover, I will review the literature on the modular organization of brain connectivity derived from neuroimaging data, and highlight the main findings and their relevance for the current models of the brain functional architecture. Finally, I will present very recent studies exploring the alterations in the modular organization of brain connectivity in patients affected by neuropsychiatric conditions, a potential marker of brain illness.

Over the past decade, numerous brain mapping techniques have been developed to investigate the integrity and pathology of structural and functional brain connectivity. Recently, the combination of multiple contrasts in Magnetic Resonance Imaging (MRI) as well as complex modeling of structural MRI data have been exploited to increase the specificity to the underlying pathological process in subjects suffering from neurological and psychiatric diseases. Applying brain mapping techniques in clinical studies requires a deep understanding of the advantages and limitations of such approaches, encompassing concepts of data acquisition, data signal processing and modeling as well as notions about the physiopathology of the targeted disease and the relevance of the clinical question. Through a number of examples, we will review how advanced MRI techniques may be used to assess changes in structural and functional properties of specific brain connections and circuits and/or of whole brain connectivity and networks. We will also focus on the advances in knowledge and limitations of such studies, concluding with future perspectives in the field of MRI mapping of brain connectivity in pathological states.

Neuronal connectivity, based on synaptic contacts, is the key structural requirement for brain functioning. The so-called brain connectome can be investigated at different levels: i) the level of "neuronal networks", involving neuronal cell populations and fiber bundles, to which imaging and computational neuroscience have especially contributed and are contributing; ii) the cellular level, focusing on the targets of axonal connectivity, whose discovery has represented a pillar in the history of neuroscience, and which has been in the focus of experimental neuroanatomy in the last decades of the twentieth century; iii) the synaptic level, aimed not only at the understanding of synapse functioning, but also at the investigation of the synaptic wiring of individual neurons or neuronal cell populations; iv) the molecular level, based, for example, on pre-and postsynaptic protein that make synaptic communication (and learning) possible. All these levels of investigations are based on different approaches and tools, providing data which need to be integrated for an overall knowledge. The lecture will go through these themes, with an emphasis on the paradigm switch imaging approaches have brought to neuroscience: the concept that functional brain connectivity, in health and disease, implements, but may not fully match, neuronal structural connectivity.