Program

June 17 (Thursday) 10:00-12:00am EST

Dynamic functional brain networks

The session explores the methodological issues in modeling dynamically changing brin networks. Chair: Hernando Ombao, KAUST, Saudi Arabia.

Liza Levina, University of Michigan, USA

Title: Network-aware neuroimaging: multiple-network analysis at the level of brain systems

Abstract: Neuroimaging technology has given us access to high-resolution measurements of connectivity between locations in the brain (nodes), resulting in high-dimensional weighted networks representing each subject. Our focus is on data from fMRI resting state data, where connectivity is typically measured as some function of correlation between time series at different nodes, and the number of nodes is typically a few hundred, resulting in tens of thousands connectivity values. When we try to predict a phenotype of interest such as disease status or a clinical assessment score using individual edge weights as features (the massive univariate approach), we can often get reasonably good predictions, but variable selection tends to be unstable and uninterpretable due to the very large number of edges. On the other extreme, reducing the entire network to a few global summaries such as modularity removes the curse of dimensionality but also loses a lot of information and does not allow for local inference. Our goal is to develop methods that use edge weights as predictors for maximum accuracy, but perform variable selection and ultimately inference at the level of brain systems, either taken from a known atlas or learned from data. Understanding the role of these systems and their interactions can lead to scientifically meaningful interpretations, unlike the edges themselves. In this talk, I will give a brief overview of two projects which aim for systems-level interpretation when predicting phenotype from individual edge features. One is on learning the most predictive systems from data, and the other on incorporating additional node covariates, such as cortical thickness, that can be measured at the same time as connectivity. The talk is based on joint work with Jesus Arroyo, Daniel Kessler, and the lab of Chandra Sripada in Psychiatry at the University of Michigan.

Chee-Ming Ting, Monash University Malaysia, Malaysia

Title: Separating stimulus-induced and background components of dynamic functional connectivity in naturalistic fMRI

Abstract: We consider the challenges in extracting stimulus-related neural dynamics from other intrinsic processes and noise in naturalistic functional magnetic resonance imaging (fMRI). Most studies rely on inter-subject correlations (ISC) of low-level regional activity and neglect varying responses in individuals. We propose a novel, data-driven approach based on low-rank plus sparse (L+S) decomposition to isolate stimulus-driven dynamic changes in brain functional connectivity (FC) from the background noise, by exploiting shared network structure among subjects receiving the same naturalistic stimuli. The time-resolved multi-subject FC matrices are modeled as a sum of a low-rank component of correlated FC patterns across subjects, and a sparse component of subject-specific, idiosyncratic background activities. To recover the shared low-rank subspace, we introduce a fused version of principal component pursuit (PCP) by adding a fusion-type penalty on the differences between the rows of the low-rank matrix. The method improves the detection of stimulus-induced group-level homogeneity in the FC profile while capturing inter-subject variability. We develop an efficient algorithm via a linearized alternating direction method of multipliers to solve the fused-PCP. Simulations show accurate recovery by the fused-PCP even when a large fraction of FC edges are severely corrupted. When applied to natural fMRI data, our method reveals FC changes that were time-locked to auditory processing during movie watching, with dynamic engagement of sensorimotor systems for speech-in-noise. It also provides a better mapping to auditory content in the movie than ISC. Joint work with Jeremy Skipper, Steven Small and Hernando Ombao.

Eric D. Kolaczyk, Boston University, USA

Title: Quantitative methods for understanding coalescence and fragmentation in dynamic networks of epileptic seizures

Abstract: Current technology for a number of measurement modalities now permits inference of dynamic brain networks over long time periods at high temporal resolution. For the study of epileptic seizures, these two attributes are likely critical for better understanding the complex process of coalescence and fragmentation of connectivity between seizure onset and termination. In this talk, I will briefly summarize two inter-related projects aimed at (i) detection of dynamic communities evolving under coalescence/fragmentation, and (ii) characterization of phase transitions underlying coalescence using a new class of random graph hidden Markov models for percolation in noisy dynamic networks.

Babak Shabbaba, University of California-Irvine, USA

Title: Scalable stochatic process modeling for dynamic brain connectivity

Abstract: We present our ongoing project aimed at understanding the neural basis of complex behaviors and temporal organization of memories. More specifically, we focus on a unique electrophysiological experiment designed to address fundamental and unresolved questions about hippocampal function. Our goal is to elucidate the neural mechanisms underlying the memory for sequences of events, a defining feature of episodic memory. To this end, we have developed a flexible Bayesian framework for dynamic modeling of brain connectivity. Using the Cholesky decomposition, our method presents the correlation matrix as the product of unit spheres with increasing dimensions. The sphere-product representation is amenable for the inferential algorithm to handle the resulting intractability, and hence lays the foundation for full flexibility in choosing priors. The proposed model, however, lacks scalability for high dimensional problems. To address this issue, we propose a latent factor Gaussian process model. We show that this approach could lead to unprecedented insight into the neural mechanisms underlying memory impairments.

June 17 (Thursday) 12:00-1:00pm EST

Plenary Talk

Michael I. Miller, Bessie Darling Massey Professor and Director of Biomedical Engineering, Co-Director of Kavli Neuroscience Discovery Institute, Johns Hopkins University, USA

Title: Hierarchical computational anatomy: towards unifying molecular and tissue scales

Abstract: Over the past several years we have been working on methods for mapping both histological and MRI scales. Towards the goal of developing a unified framework for representing multi-scale particle and tissue anatomy and function we model the brain as a layered hierarchy of measures. Hierarchical mapping occurs as a series of flows of diffeomorphisms across layers, with the vector fields successively refined connecting the coarse tissue scale motions with the fine scale particle motions. As part of the representation we present a formalism for scale-space aggregation that traverses the layers of by projecting functional descriptions across scales into stable empirical probability laws. Hierarchical measure norms are used for mapping the layers simultaneously. We examine several examples in Alzheimer’s histology and MR imaging as well as spatial transcriptomics.

June 17 (Thursday) 1:00-2:00pm EST

Break & Social Interaction

June 17 (Thursday) 2:00-3:30pm EST

Brain Development and Imaging Genetics

The session explores methodical development in brain development and imging genetics. Chair: Anqi Qiu, National University of Singapore.

Gang Li, University of North Carolina, USA

Title: Baby brain mapping with machine learning

Abstract: The increasing availability of pediatric brain MRI data affords unprecedented opportunities for precise charting of dynamic early brain developmental trajectories in understanding normative and aberrant brain growth. However, conventional neuroimaging computational tools developed for adults are inapplicable for processing the challenging pediatric MR images, due to their extremely low tissue contrast and regionally-heterogeneous dynamic changes of imaging appearance, brain structure and function. In this presentation, I will introduce our developed deep learning techniques for addressing critical steps in quantitative mapping of baby brain development, e.g., cortical surface correction, parcellation, registration, harmonization, and atlas construction. In particular, to respect and leverage the inherent spherical topology of the cerebral cortex, we have created novel spherical neural networks for learning complex cortical data on the non-Euclidean space.

Wei Gao, Department of Biomedical Sciences, Cedars-Sinai, USA

Title: New Concept of Brain-Behavior Relationship Heterogeneity and Implications/Adaptations in Novel Brain-Based Prediction Models

Abstract: Most conventional brain-behavior association studies assume homogeneity within a particular studied population. Deviating from this assumption, we have examined a novel concept of relationship-level heterogeneity in a cohort of otherwise “homogeneous” typically developing newborns and demonstrated the existence of two subgroups with contrasting brain-behavioral mechanisms. Such heterogeneity in brain-behavioral relationships inevitably creates challenges for conventional brain-based prediction models of behavioral outcomes since a whole-group model might not work equally well for the different subgroups with contrasting brain-behavioral associations. In responding to this challenge, we derived a novel “brain outlier-based” prediction model, named “Triple-Outliers (Triple-O)” that is not sensitive to such relational heterogeneity for prediction and does not need training with behavioral outcome data for improved robustness and generalizability. Our initial empirical results showed descent level of sensitivity, high level of specificity and overall accuracy, as well as high level of robustness/generalizability in using newborn brain functional connectivity features to predict 4-year IQ outliers based on Triple-O. I will describe the related methods, findings, and implications for future directions in this talk. Link to talk

Li Shen, University of Pensilvannia, USA

Title: Brain imaging genetics for Alzheimer's disease: integrated analysis and machine learning

Abstract: Brain imaging genetics is an emerging data science field, where integrated analysis of brain imaging and genetics data, often combined with other biomarker, clinical and environmental data, is performed to gain new insights into the genetic, molecular and phenotypic characteristics of the brain as well as their impact on normal and disordered brain function and behavior. Many methodological advances in brain imaging genetics are attributed to large-scale landmark biobank projects such as the Alzheimer’s Disease Sequencing Project, the Alzheimer’s Disease Neuroimaging Initiative, and the UK Biobank. Using the study of Alzheimer’s disease as an example, we will discuss fundamental concepts, state-of-the-art statistical and machine learning methods, and innovative applications in this rapidly evolving field. We show that the wide availability of brain imaging genetics data from various large-scale biobanks, coupled with advances in biomedical statistics, informatics and computing, provides enormous opportunities to contribute significantly to biomedical discoveries in brain science and to impact the development of new diagnostic, therapeutic and preventative approaches for complex brain disorders such as Alzheimer’s disease.

Yalin Wang, Dept. of Computer Science, Arizona State University, USA

Title: Quantitative Characterization of the Human Retinotopic Map Based on Quasiconformal Mapping

Abstract: The retinotopic map depicts the cortical neurons' response to visual stimuli on the retina and has contributed significantly to our understanding of human visual system. Although recent advances in high field functional magnetic resonance imaging (fMRI) have made it possible to generate the in vivo retinotopic map with great detail, quantifying the map remains challenging. Existing quantification methods do not preserve surface topology and often introduce large geometric distortions to the map. We introduced a new framework based on computational conformal geometry and quasiconformal Teichmuller theories to quantify the retinotopic map and further refine its topological condition. Specifically, we adopted the Beltrami coefficient, a metric of quasiconformal mapping, to rigorously and complete characterize the retinotopy map by quantifying the local quasiconformal mapping distortion at each visual field location. The obtained Beltrami coefficient map (BCM) provided fully invertible retinotopic maps. We successfully applied the new framework to analyze the V1 retinotopic maps from the Human Connectome Project (HCP, n=181), the largest state of the art retinotopy dataset currently available. With unprecedented precision, we found that the V1 retinotopic map was quasiconformal and the local mapping distortions were similar across observers. The new framework can be applied to other visual areas and retinotopic maps of individuals with and without eye diseases, and improve our understanding of visual cortical organization in normal and clinical populations. In this talk, we will also briefly introduce our recent progress on retinotopic map topological smoothing and retinotopic map registration research. Link to talk

June 17 (Thursday) 4:00-4:30pm EST

Offline prerecorded session

This new session introduces few prerecorded talks that can be viewed at your own time. Brief introduction to each talk will be given. Chair: Moo K. Chung, University of Wisconsin-Madison. The offline prerecorded videos are archived in https://www.youtube.com/channel/UC_48A5VaPPXJ6q3vxXKGp-A

Yu-Ping Wang, Dept. of Biomedical Engineering, Tulane University, USA

Title: Fusion of multimodal brain connectivity networks for IQ prediction with alternative diffusion map

Abstract: Functional connectivity (FC) has been used to study individual differences in development, behavior, and cognition. However, current approaches are mainly using linear dimensionality analysis for extracting essential network patterns, which fail to capture the nonlinearity of the brain network. Herein, we propose a framework based on alternating diffusion map (ADM) to extract geometry-preserving low-dimensional embeddings. Specifically, we first separately build resting-state and task-based FC networks by symmetric positive definite matrices using sparse inverse covariance estimation for each subject, and then utilize the ADM for their fusion. Finally, the low-dimensional embeddings are extracted as fingerprints to identify individuals. The proposed framework is validated on the Philadelphia Neurodevelopmental Cohort data for the classification of intelligence quotient (IQ). The fusion of resting-state and n-back task fMRI by the proposed framework achieves better classification accuracy than any single fMRI, and the proposed framework is shown to outperform several other data fusion methods. Link to talk

Doug Dean III, University of Wisconsin-Madison, USA

Title: Quantitative Imaging of the Developing Brain

Abstract: The development of the brain is a lifelong process that results in a remarkable transformation of the neural architecture; however, this change is perhaps most rapid during the first years of life. Processes that are fundamental to brain connectivity and that help facilitate the advancement of higher-level cognitive functioning, undergo a rapid and pronounced pattern of development that is accelerated from birth to 5 years. Moreover, the neural substrates that govern individual differences toward vulnerability or resilience to adversity likely develop during this period, making neurodevelopmental processes susceptible to early experiences and alterations. Quantitative magnetic resonance imaging makes it possible to acquire measurable and reproducible quantities that can provide insight into the underlying processes of early brain development as well as allow one to begin to explore emerging relationships between cognition and developmental disability. In this talk, I will highlight recent advancements that have made it possible to perform MRI in infant and young children populations as well as describe how quantitative MRI techniques are being used to gain a better understanding of the highly dynamic and nonlinear mechanisms that support early brain development. Link to talk

Ilwoo Lyu, Ulsan National Institute of Science and Technology (UNIST), Korea

Title: Improving cortical surface annotation by spherical data augmentation using non-rigid deformation

Abstract: Regional-based morphological analysis is a widely adapted approach in neurodevelopmental studies. This necessitates consistent subdivision of cortical surfaces into multi-regions based on cortical parcellation protocols in an anatomical or functional fashion. Yet, consistent labeling of cortical regions is challenging due to the complicated cortical folds and inter-subject variability. A promising way forward to achieve this goal is convolutional neural networks (CNNs) that have recently shown remarkable achievement in semantic segmentation over traditional machine learning techniques. CNNs generally tend to improve performance as training samples increase; nevertheless, the acquisition of surface data annotation is very expensive, which requires neuroanatomical expertise that most human brain mappers do not have. In this talk, we present cortical surface data augmentation to amplify the variations within training samples. To take account of neuroanatomical variability, training samples are synthesized from the proposed feature space that embeds intermediate deformation trajectories between a pair of training samples in a rigid to non-rigid manner, which bridges an augmentation gap in conventional rotation data augmentation. The proposed method is evaluated on two labeling tasks: whole brain parcellation and sulcal basin labeling in the lateral prefrontal cortex. In the latter task, a two-stage training process is employed to improve labeling accuracy of tertiary sulci by informing the biological associations to primary/secondary sulci. The experimental results show that the proposed method improves labeling accuracy over the baseline techniques. Link to talk

Eardi Lila, Dept. of Biostistics, Univeristy of Washington, USA

Title: Joint modeling of brain shape and functional connectivity

Abstract: Human brains display both anatomical and functional variations across subjects. However, population analyses of brain images are usually performed following a spatial normalization step, which has the effect of discarding anatomical variability, leading to models that can only provide limited scientific insights. We, therefore, propose a novel statistical framework for the analysis of associations between subject-specific anatomical and functional connectivity configurations. The proposed framework is moreover able to disentangle associations that are due to genetic sources from those driven by unique environmental factors. Finally, we apply the proposed model to the Human Connectome Project dataset to explore spontaneous co-variation between brain shape and connectivity in young healthy individuals. Link to talk

Yuan Wang, Dept. of Epidemiology & Biostatistics, University of South Carolina, USA

Title: Topological Signal Processing and Inference on Event-Related Potential Response

Abstract: Topological signal processing decodes structural changes in signals through persistent homology. Group comparison of topological signal features requires reliable statistical inference procedures. Having advanced a permutation-based framework on persistence landscape (PL) in single-trial electroencephalography (EEG), we extend it to event-related potential (ERP) response through an exact permutation test on PLs. The new framework is applied to identify spatial neural deficits in post-stroke aphasia associated with vocalization in an altered auditory feedback (AAF) study with simultaneous EEG recording. Link to talk

Won Hwa Kim, Dept. of Computer Science, POSTECH, Korea

Title: Multi-resolution Edge Network (MENET) for Alzheimer’s Disease Classification with Brain Network

Abstract: Tremendous literature indicate that Brain Networks, defined by associations between different regions of interests (ROIs) in the brain, show early signs of neurodegenerative diseases. Brain networks are naturally represented as graphs that consists of sets of nodes and edges. Their irregular structures differentiate the graphs from traditional imaging data in Euclidean spaces, and thus it requires sophisticated machine learning and signal processing techniques for analysis. In this talk, I will introduce our recent effort on analyzing brain networks in a multi-resolution fashion and report scientific findings that identify disease-specific variations in the brain due to Alzheimer’s Disease (AD). Our framework is sensitive enough to capture even subtle changes in the early stages of AD outperforming conventional approaches. Link to talk

June 18 (Friday) 10:00-12:00am EST

Interaction of geometry and topology

The session expolores the geometric and toplogical issues arising in brin imaging data and models. The explcit exploitation of underlying geometry and topology can increase the performance. Chair: Moo K. Chung, University of Wisconsin-Madison, USA.

Maxime Descoteaux, Dept. of Computer Science, Université de Sherbrooke, Canada

Title: Surface geometry to enhance tractography & improve structural connectivity

Abstract: In this talk, methods that combine cortical surface meshes with tractography reconstruction are presented to improve endpoint precision and coverage. Mapping diffusion MRI tractography streamlines to the cortical surface facilitates the integration of white matter features onto gray matter, especially for connectivity analysis. This mapping also enables the study of structural measures from tractography along the cortex and subcortical structures. In addition to structural connectivity analysis, novel adaptive and dynamic surface seeding methods are proposed to increase the cortical coverage and to reduce endpoint location biases. Use of cortical and subcortical meshes together with a proper seeding strategy reduces the variability in structural connectivity analysis. Talk Link

Jong Chul Ye, Dept. of Bio and Brain Engineering, KAIST, Korea

Title: Physics-informed Unsupervised QSM Deep Learning using Optimal Transport CycleGAN

Abstrct: Quantitative susceptibility mapping (QSM) is a useful magnetic resonance imaging (MRI) technique which provides spatial distribution of magnetic susceptibility values of brain tissues. QSMs can be obtained by deconvolving the dipole kernel from phase images, but the spectral nulls in the dipole kernel make the inversion ill-posed. In recent times, deep learning approaches have shown a comparable QSM reconstruction performance as the classic approaches, despite the fast reconstruction time. Most of the existing deep learning methods are, however, based on supervised learning, so matched pairs of input phase images and the ground-truth maps are needed. Moreover, it was reported that the supervised learning often leads to underestimated QSM values. To address this, here we propose a novel unsupervised QSM deep learning method using physics-informed cycleGAN, which is derived from optimal transport perspective. In contrast to the conventional cycleGAN, our novel cycleGAN has only one generator and one discriminator thanks to the known dipole kernel. Experimental results confirm that the proposed method provides more accurate QSM maps compared to the existing deep learning approaches, and provide competitive performance to the best classical approaches despite the ultra-fast reconstruction.

Shantanu H. Joshi, Department of Neurology, University of California - Los Angeles, USA

Title: A Differential Geometric Approach for fMRI Time Series Alignment

Abstract: This talk presents a differential geometric approach for alignment of functional magnetic resonance time series data. We achieve temporal alignment of both amplitude and phase of the functional magnetic resonance imaging (fMRI) time course and spectral densities. Experimental results show significant increases in pairwise node to node correlations and coherences following alignment. Additionally, we show results for task based fMRI signals, where we see improved power of detection of clusters and activations for single subject data. We also present a geometric approach for minimizing the variability in the shape of along- tract diffusion profiles by performing diffeomorphic alignment across the tracts as well as across populations. Finally, we present an approach for accelerating the alignment process using deep learning.

Moo K. Chung, University of Wisconsin - Madison, USA

Title: Topological inference and learning for brain networks

Abstract: Many previous studies on brain networks have mainly focused analyzing graph theory features that are often parameter dependendent. Persistent homology provides a more coherent mathematical framework that is invariant to the choice of parameters. Instead of looking at networks at a fixed scale, persistent homology charts the topological changes of networks over every possible scale. In doing so, it reveals the most persistent topological features that are robust to scale changes. In this talk, we present a novel topological infernce and learning framework that can integrate networks of different sizes, topology or modalities through persistent homology. This is possible through the Wasserstein distance that measures the topoloical similarity between networks. The use of Wasserstein distance on graph filtrations bypasses the intrinsic computational bottleneck associated with matching networks. This talk is based on Chung et al. 2019 (Network Neuroscience 3:674-694) and preprint arXiv:2012.0067.

June 18 (Friday) 12:00-2:00pm EST

Deep learning in brain imaging

The session explores deep learning methods used in brain imaging Chair: Vince Calhoun, Georgia State University, USA

Satrajit Ghosh, Masschusetts Institute of Technology, USA

Title: Nobrainer: A deep learning toolkit for neuroimaging

Abstract: Deep learning technology has exploded in recent years, dramatically improving the state-of-the art performance for applications in many domains. In neuroimaging, the amount of available data is increasing yearly through several international brain initiatives and biobanks. Significant improvements in hardware has supported a rapid increase in deep learning models alongside this growth in data. However, progress in the field has been impeded by limited availability of well-tested models and ready to use applications that have been trained on large and diverse datasets. In this talk, I will describe the Nobrainer framework, its goals to build a community resource, and current state of the art models and applications available for use. I will discuss the use of Bayesian neural networks for more informed prediction and distributed learning, the development and use of generative models, and current adventures in limiting catastrophic forgetting. Finally, I will situate the framework in the context of a larger community-based neuroimaging ecosystem of public and private partnerships to improve access to diverse data, to increase robustness of model training, and to highlight real-world challenges that need to be addressed.

Nikolaus Kriegescorte, Columbia University, USA

Title: Testing deep neural network models of human vision with neuroimaging data

Abstract: To learn how cognition is implemented in the brain, we must build computational models that can perform cognitive tasks, and test such models with brain and behavioral experiments. Neural network models have enabled major strides in computer vision and other artificial intelligence applications. This brain-inspired technology provides the basis for tomorrow’s computational neuroscience. Deep convolutional neural nets trained for visual object recognition have internal representational spaces remarkably similar to those of the human and monkey ventral visual pathway. Functional imaging and invasive neuronal recording provide rich brain-activity measurements in humans and animals, but a challenge is to leverage such data to gain insight into the brain’s computational mechanisms. I will discuss statistical inference techniques that enable us to adjudicate among deep neural network models on the basis of neuroimaging data. Recurrent convolutional neural networks also provide better accounts of the dynamics of human ventral-stream visual representations, as measured with magnetoencephalography (MEG). Current models still fall short of explaining how humans can so rapidly, robustly, and deeply understand the causes and implications of a visual image. However, the existing tools of measurement and modeling and the emerging methods for testing models with measurements are accelerating progress in cognitive computational neuroscience.

Weizheng Yan, TReNDS Center, USA

Title: Deep Learning applications in fMRI: classification, multi-view fusion and subtype discovery.

Abstract: Deep learning technology has been applied to broad areas including image/video recognition and neuroimaging. However, compared to natural images, neuroimaging data is more complex, usually with higher dimensions, smaller sample size, multiple data types (e.g., brain structure and function), and often lacking a solid ground truth. This presentation will discuss three categories of deep learning applications in fMRI including 1) a multi-scale convolutional recurrent neural network for leveraging spatio-temporal dynamic information; 2) a multi-view fusion strategy to combine deep-learning and standard machine learning for fMRI classification; 3) a unified framework for supervised classification and unsupervised subtype clustering of psychotic disorders.

Baiyang Lei, School of Biomedcial Engineering, Shenzhen University, China

Title: Multi-modal and Multi-time Neuroimaging Deep Learning for Brain Disease Analysis

Abstract: Neuroimaging is safe and non- invasive, which is widely used in the early diagnosis and prediction of Alzheimer's disease, Parkinson's disease, Autism and other brain diseases. With the rapid development of intelligent computing technology, it is possible to use neuroimaging data for intelligent diagnosis of brain diseases. Starting from the clinical needs of brain diseases, we have carried out the research on intelligent diagnosis and prediction of brain diseases through multi-modal and multi- time point neuroimage data, and has achieved some preliminary results, which has become a complete research system and technical chain. Guided by the practical problems in clinical diagnosis, this paper proposes a deep learning algorithm for neuroimaging data, and puts the final research results into clinical practice. The report focuses on machine learning, deep learning and data mining algorithms in the field of computer vision, and carries out systematic research on the early diagnosis of brain diseases: 1) a series of feature learning algorithms are proposed to solve the problems of high specificity, large individual differences and high dimension of neuroimaging in patients with Alzheimer's disease, Parkinson's disease, Autism and other brain diseases; 2) To explore the construction of brain network and explore the internal relationship between brain function degradation and brain activation;3) based on deep learning and machine learning, the early diagnosis model of brain diseases is established to improve the accuracy of diagnosis and prediction.

June 18 (Friday) 2:00-3:00pm EST

Break & Social Interaction

June 18 (Friday) 3:00-5:00pm EST

Functional brain networks

The session explores the methodogical issues in modeling dynamically changing brin networks. Chair: Martin Lindquest, Jonhs Hopkins University, USA

Richard Betzel, Psychological and Brain Sciences, Indiana University, USA

Title: Edge-centric connectomics

Abstract: Network neuroscience has relied on a node-centric network model in which cells, populations, and regions are linked to one another via anatomical or functional connections. Other disciplines, however, have prioritized network edges and created edge-centric network representations leading to new insights about system organization and function. Recently, we developed an edge-centric network model that generates novel constructs of ‘edge time series’ and ‘edge functional connectivity’ (eFC). In my talk I will summarize the results of several recent papers and highlight some of the advantages that edge-centric network models have over traditional node-centric network representations. These include an exact decomposition of functional connectivity into its framewise contribution and the ability to resolve overlapping system-level architecture.

Brian Caffo, Johns Hopkins University, USA

Title: Covariance regression for connectome outcomes

Abstract: In this talk, we cover methodology for jointly analyzing a collection of covariance or correlation matrices that depend on othervariables. This covariance-as-an-outcome regression problem arises commonly in the study of brain imaging, where the covariance matrix in question is an estimate of functional or structural connectivity. Two main approaches to covariance regression exists: outer product models and joint diagonalization approaches. We investigate joint diagonalization approaches and discuss the benefits and costs of this solution. We distinguish between diagonalization approaches where the eigenvectors are selected in the absence of covariate information and those that chose the eigenvectors so that the result regression model holds best. The methods are applied to resting state functional magnetic resonance imaging data in a study of aphasia and potential interventions.

Mandy Mejia, Department of Statistics, Indiana University, USA

Title: Using empirical population priors to understand functional topology and connectivity in individuals

Abstract. A primary objective in resting-state fMRI studies is localization of functional areas (i.e. resting-state networks) and the functional connectivity (FC) between them. These spatial and temporal properties of brain organization may be related to disease progression, development, and aging, making them of potentially great scientific and clinical value. A common tool to estimate functional areas and their FC is independent component analysis (ICA). Due to high noise levels in fMRI and typically short scan durations, subject-level ICA results tend to be highly noisy and unreliable. Thus, group-level functional areas are often used in lieu of subject-specific ones, ignoring inter-subject variability in functional topology. These group-average maps also form the basis for estimating FC, leading to potential bias in FC estimates. An alternative to these two extremes (noisy subject-level ICA and one-size-fits-all group ICA) is Bayesian hierarchical ICA, wherein information that is shared across subjects is leveraged to improve subject-level estimation of spatial maps and FC. However, fitting traditional hierarchical ICA models across many subjects is computationally intensive. Template ICA is a hierarchical ICA framework using empirical population priors to provide the benefit of hierarchical ICA with fast computation. Template ICA allows for more accurate estimation of subject-level functional areas and the functional connectivity between them. Two extensions involve spatial priors and empirical population priors on functional connectivity. Template ICA has been used in a range of applications, including a study of the effects of psilocybin (the prodrug compound found in “magic mushrooms”) on the organization of the thalamus and cortico-thalamic connectivity.

Robyn Miller, Tri-institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS), USA

Title: Evolving connectivity motifs (EVOdFNCs) and functionally localized entropy in the dynamic connectome: new methods and new insights into resting-state brain dynamics

Abstract. Despite growing interest in time-varying functional network connectivity (dFNC), much analytical innovation has continued to revolve around a hidden Markov paradigm applied at the whole-brain level. The focus is typically on the identifying a small collection of fixed, i.e,. non-varying, patterns of functional organization measured on timescales shorter than that of the full scan, then characterizing the high-dimensional whole-brain connectivity dynamics as a memoryless Markov process on this discrete set of indexed states. This appealing simplification has potentially consequential limitations for modeling functional connectivity as effective mental functioning in a complex, unpredictable, multifaceted world requires intricate memory-laden contingencies. The development of analytical tools that capture valid, naturalistic, group-level representations of the fluidly evolving resting-state connectome is a challenging, and arguably crucial, next step toward fully leveraging the information provided by large functional imaging studies. We will discuss a new data-driven framework for identifying “movie-style” multiframe temporally evolving dynamic functional network connectivity states (EVOdFNCs). Applied to data from a large rs-fMRI study of schizophrenia (SZ), we find that patients and controls exhibit different dynamics around common transiently realized whole-brain connectivity states, and that those patients with higher levels of key symptom are more distinguishable from other SZ patients under this dynamic lens than they are using the standard HMM approach. Time-permitting we will also discuss an integrated local-global approach to investigating localized patterns of functional coherence within a whole-brain hidden Markov framework. Applied to a large resting-state fMRI schizophrenia study, this analysis reveals that functionally localized entropy (FLE) distributes differently over the connectome in patients and controls, varies significantly based on which HMM state the corresponding whole-brain dFNC observation is occupying and also based on which HMM transition the dFNC observation is poised to make. Patients exhibit directional effects on the distribution of FLE that coincide with HMM effect patterns on timepoints preceding transitions to target HMM states more occupied by SZs. This suggests localized entropic patterns may be a latent factor disposing patient dFNC observations toward whole-brain HMM clusters that SZ patients have been consistently shown to spend more time in than controls. The functioning human brain must react, adapt, learn and reorganize over many functional and temporal scales. Our work underscores the potential benefits of gaining methodological access to a richer array of these scales in functional imaging data. Link to talk