The HaxbyLab@Dartmouth

Neural Computation: Computational Approaches to Brain Connectivity (Abstracts)


Neural Computation: Computational Approaches to Brain Connectivity (Abstracts)

Please return to the main page of the workshop for more information.


Computational approaches

Herve Abdi (University of Texas at Dallas)

Multi-Table Models for Connectivity Analysis (Parts One and Two)

Stephen Strother (University of Toronto)

The Ugly, the Bad and the Good of Predictive Modeling for Functional Brain Networks

An important goal for predictive models of BOLD fMRI data is to discover the distributed neural networks supporting task performance, and to validate the resulting spatial brain patterns using known network structures and/or links with behavioral responses. I will address this goal in three different multivariate predictive modeling scenarios and emphasize the importance of the tradeoff that exists in such analyses between prediction accuracy and spatial pattern reliability. The Ugly: The first scenario uses 14 subjects performing a bilateral finger-tapping task with whole-brain predictive models based on Support Vector Machines, Logistic Regression and Fisher’s Discriminant Analysis. I will show that prediction alone and/or prediction combined with sparse spatial pattern constraints (e.g., elastic net or recursive feature elimination), without consideration of spatial pattern reliability, leads to network patterns with significant false negatives. The Bad: The second scenario is an extreme example of a prediction vs spatial reliability tradeoff with minimized prediction (i.e., no better than guessing) and maximized spatial reliability. Using a multi-task, block activation study in 19 subjects I will show that this tradeoff may be used to discover strong links between behavioral performance and a brain pattern reflecting oscillation between default-like vs task-positive brain networks during task performance. The Good: The third scenario uses 26 subjects performing the Trail-Making task in the scanner, and focuses on extracting the neural basis of the cognitive difference between Trail B vs Trail A tasks. I will show that optimization of pre-processing pipeline steps in individual subjects can dramatically boost prediction of behavioural responses across subjects, and that prediction is an important constraint for this optimization.

Pierre Bellec (University of Montreal)

Multiscale and multilevel bootstrap analysis of stable clusters in fMRI

Functional MRI gives an indirect measure of brain activity with good spatial and temporal resolution. Clustering techniques can be used to summarize such large datasets into brain networks, i.e. groups of brain regions with similar temporal activity. The cluster analysis of an fMRI database raises (at least) two challenges. First, it has long been hypothesized that the brain follows a modular organization where large systems (such as visual, motor, etc) can be further decomposed into subnetworks with an increasing level of functional specialization. The cluster analysis thus needs to be performed at multiple scales (i.e. number of networks). Second, even after a perfect anatomical coregistration, the functional cerebral architecture will vary from subjects to subjects. It is thus necessary to find a multi-level clustering model that is flexible enough to capture individual features while finding a consistent solution across subjects (so that individual cluster maps can be compared at the group level). In this talk I will discuss a technique called bootstrap analysis of stable clusters (BASC) [BPJ+06] [BRL+10], which can be used to perform multilevel and multiscale analysis of resting-state fMRI. I will also present new results showing how this technique applies in the context of task-based fMRI and how it can be used to optimize the detection power in a connectome-wide association study.

[BPJ+06]P. Bellec; V. Perlbarg; S. Jbabdi; M. Pelegrini-Issac; J.L. Anton; H. Benali, Identification of large-scale networks in the brain using fMRI. Neuroimage, 2006, 29
[BRL+10]P. Bellec; P. Rosa-Neto; O.C. Lyttelton; H. Benalib; A.C. Evans, Multi-level bootstrap analysis of stable clusters in resting-State fMRI. Neuroimage 51 (2010), pp. 1126-1139.

Resting state functional connectivity

Steve Petersen (Washington University)

Koene R.A. Van Dijk (Harvard University)

Resting State Functional Connectivity: Promises and Pitfalls

First I will present evidence indicating that we are able to reliably measure the architecture of large-scale neuronal systems in the human brain using resting state functional connectivity MRI (fcMRI). Then I will touch upon the topic of confounding effects of head motion on fcMRI metrics and continue with an example of one of our studies in which we used multi-variate pattern analysis (MVPA) on resting state fMRI data to predict advanced aging. Finally, I will show some preliminary data indicating the advantage of ultra-high spatial resolution fcMRI at 7 Tesla.

Brain-to-brain connectivity

Uri Hasson (Princeton University)

Inter-subject functional connectivity: a new tool for exploring the mechanisms of dyadic social interactions

Cognition materializes in an interpersonal space. The emergence of complex behaviors requires the coordination of actions among individuals according to a shared set of rules. Despite the central role of other individuals in shaping our minds, experiments typically isolate human or animal subjects from their natural environment by placing them in a sealed quiet room where interactions occur solely with a computer screen. In everyday life, however, we spend most of our time interacting with other individuals. In the talk I will argue in favor of a shift from a single-brain to a multi-brain frame of reference. I will present a new analysis tool, in which we compute the ?functional connectivity? between the brain responses in a seed area in one subject and the responses in other subjects? brains. While at rest we see no correlations in the responses across subjects, during the processing of real life stimuli the brain responses in one brain are coupled to the responses in another brain. Such neural coupling is mediate via the transmission of a signal (stimulus-to-brain coupling) through the environment. When the transmitted signal is speech signal which was produced by another brain, the inter-subject functional analysis exposes a shared neural substrate that exhibits temporally aligned response patterns across the speaker and the listener. The recording of the neural responses from two brains opens a new window into the neural basis of interpersonal communication, and may be used to assess verbal and non-verbal forms of interaction in both human and other model systems.

Diffusion imaging and tractography

Christian Beaulieu (University of Alberta)

The KISS principle applied to diffusion tensor tractography of the human brain

Diffusion magnetic resonance imaging and tractography, in its many forms, has made major advances for the non-invasive study of human brain white matter wiring in unprecedented ways. While the tensor model has fallen out of favour given more complex acquisition and post-processing schemes, it still provides an informative framework for many studies. The goals of the talk are: (a) to introduce the basic concepts of diffusion imaging, (b) to provide an overview of using tractography to define regions-of-interest for interrogating the brain, (c) to discuss links between tensor-derived diffusion parameters and the underlying tissue micro-structure by highlighting several post-surgical studies, (d) to provide an overview of DTI-tractography small world networks, and (e) to illustrate the advantages and considerations of diffusion imaging at high static fields such as 4.7 Tesla.

Yu-Chien Wu (Dartmouth)

Diffusion MRI and Fiber Orientation Function

In this talk, I’ll first introduce general diffusion imaging techniques. I’ll start from the diffusion physics and then discuss diffusion tensor imaging (DTI), high angular resolution diffusion imaging (HARDI) and q-space imaging (QSI). Lastly, I’ll focus on the fiber orientation distribution functions for white matter (WM) tractography.

Carl-Fredrik Westin (Harvard University)

Novel opportunities in diffusion MRI

The diffusion MRI (dMRI) technique has raised hopes in the neuroscience community for a better understanding of the white matter anatomy of the human brain. The hope is that the extension of available technology will aid in the diagnosis and subsequent treatment of disorders of the central nervous system and is likely to have a major impact on assessment of white matter pathologies (e.g., schizophrenia, multiple sclerosis), detection of stroke and trauma including traumatic brain swelling, diffuse axonal injury, and spinal trauma, as well as a large variety of brain tumors. In this talk I will review two recent developments in dMRI: 1) compressed sensing, and 2) double pulsed field gradient (double-PFG) dMRI, and discuss opportunities from these new technologies. I will also review results from last year’s diffusion tractography challenge workshop at MICCAI 2011, illuminating opportunities and challenges for mapping white matter pathways in-vivo.

Saad Jbabdi (Oxford University)

Recent developments in data acquisition, pre-processing and modelling of diffusion MRI data for human connectomics

Diffusion tractography offers great potential for the study of human brain anatomy. However, as a method to study brain connectivity, tractography suffers from serious limitations. In this talk, I will present some recent developments, made by the Human Connectome Project consortium, that are aimed to address some of these limitations. These advances span data acquisitions, pre-processing and voxel-wise modelling of the diffusion data.

General discussion

Barry Horwitz (NIDCD)

Overview of Computational Approaches to Brain Connectivity

I will touch on the main themes of the workshop, mentioning some of the topics that merit further discussion. Also, I will bring up some issues related to brain connectivity that future research will have to address. Given that the three components of a network are its nodes, its inter-nodal links, and the inter-nodal connection weights, I will mention some approaches for better defining the links and nodes of brain networks. Given the dynamically nature of brain networks, I will discuss the importance of dealing with the changing temporal relationships between network components. Finally, I will briefly discuss how neural modeling can be used to help interpret brain connectivity data