The pdf of the posters' abstracts can be downloaded by clicking on this link
1. Asish Banik, Andrew Bender, Tapabrata Maiti
Michigan State University, Michigan State Uinversity, Michigan State University
Bayesian Spatiotemporal clustering model for analyzing White Matter Data
We present a new approach of Bayesian model based clustering for Spatiotemporal data in this chapter. An extension of Linear mixed effects modeling is applied for spatiotemporal cerebral white matter data extracted from healthy aging individuals. LME provides us a prior information for spatial covariance structure and brain segmentation based on white matter intensity. This motivates us to build a stochastic modelbased clustering to group voxels considering their longitudinal and location information. The cluster specific random effect brings correlation among repeated measures. The problem of finding partitions is dealt with imposing prior structure on cluster partitions. Finally an objective function is derived to find optimal clusters in brain. Optimization of the objective function is the most challenging task for this problem due to high number of voxels present in brain. We implemented a Metropolis-Hastings algorithm for detecting partitions. Moreover, we tried to use penalized approach in order to select voxels associated with fixed effects. We will present the results over various clusters or regions of brain and how different estimates differ from one region to another.
Keywords: Model based clustering, stochastic search, linear mixed effect model, Bayesian methods, objective score function, white matter data
2. Joanne C. Beer, Russell T. Shinohara, Kristin A. Linn*
Perelman School of Medicine, University of Pennsylvania
Harmonization of multi-scanner longitudinal MRI neuroimaging data
Aggregation of neuroimaging datasets from multiple sites and scanners is becoming increasingly common. While this presents opportunities for increased statistical power, it also presents challenges due to systematic scanner effects. We propose a method for the harmonization of multi-site longitudinal MRI data based on ComBat, a method originally developed for genomics and later adapted to cross-sectional MRI data. In simulation studies, we assess the statistical properties of longitudinal ComBat. Using longitudinal cortical thickness data from the Alzheimer's Disease Neuroimaging Initiative (ADNI), we demonstrate the presence of scanner-specific location and scale effects. We compare estimates of the association between baseline diagnosis group and change in cortical thickness over time using three versions of the ADNI data: (1) raw data, (2) data harmonized using cross-sectional ComBat, and (3) data harmonized using longitudinal ComBat.
3. Emilie Campos, Damla Senturk, Chad Hazlett
Principle ERP reduction: Estimating and analyzing principle ERP waveforms underlying ERPs across subjects, channels, and conditions
ERP waveforms are the summation of many unknown overlapping signals. Due to this summation, changes in the peak or mean amplitude of a waveform over a given time period cannot be attributed to a particular ERP component of ex ante interest, as is the standard approach to ERP analysis. Though this problem is widely recognized, solutions have remained out of reach and the problem is largely neglected in practice. Our approach begins with the presumption that any observed ERP waveform ‚Äì at any electrode, for any trial type, and for any participant or groups of participants ‚Äì is a weighted combination of signals from an underlying set of what we refer to as principle ERPs, or pERPs. With a dataset spanning multiple tasks and participants the principle ERP reduction (pERP-RED algorithm) estimates a set of principle ERPs. Investigators can then translate any observed waveform (for a given group or individual on a given condition) into a vector of coefficients describing the amplitude of each of the pERPs that contribute to it. Among other potential uses, any contrast or comparison that would normally be conducted with ERPs can instead be conducted on these amplitude measures. Users can also visually inspect the principle components implicated in a given contrast to better understand their time course, as well as their scalp distribution. We demonstrate the pERP-RED algorithm and subsequent application through extensive simulations and on real data collected from experiments on participants diagnosed with Autism Spectrum Disorder (ASD) and Attention Deficit Hyperactivity Disorder (ADHD).
Keywords: event related potentials, principal components analysis and independent component analysis
4. Naomi Kaplan Damary
Forensic Footwear Analysis‚ The Location Of Randomly Acquired Characteristics (Racs )
Forensic experts use randomly acquired characteristics (RACs) on the shoe sole, such as scratches or holes, in order to compare suspects' shoes with prints found at the crime scene. This paper presents methods of estimating the probability that a RAC appears at a given location on a shoe sole, a preliminary question in determining the shoe's degree of rarity and ultimately its evidential value in the court room. An extensive data base of real suspects' shoe prints collected recently by the Israel National Police (Yekutieli et al., 2016) is used to examine the distribution of RAC locations. RACs are assumed to follow a two-dimensional point process, the intensity function of which governs the probability of observing RACs in different areas of the shoe sole. The use of real shoe data presents various challenges which are dealt with here through a model suggested for the intensity function. Three estimators of this function are compared using two types of analyses. The first, which is based on a pixel analysis using a logistic regression and natural cubic splines, creates a smooth intensity function. Due to computation challenges, case-control sub-sampling techniques are used. This method, however, has some limitations which a second form of analysis, based on larger areas and a piece-wise constant intensity function, helps resolve, in spite of its lower resolution. It was found that RACs were more than twice as likely to appear in certain locations than in others, apparently due to the morphology of the foot which exerts pressure on the shoe.
Keywords: Accidental marks, Case-control sampling, Conditional maximum likelihood, Footwear comparison, Natural cubic splines, RACs, Random effects model, Randomly acquired characteristics, Shoe pattern, Shoe print.
Tianyu Ding , Annie Cohen , Erin O'Connor , Helmet Karim , Lea Alhilali
, Adina Crainiceanu , John Muscheli , Oscar Lopez , William Klunk ,
Howard Aizenstein , Rob Krafty , Ciprian Crainiceanu , Dana Tudorascu
University of Pittsburgh, University of Maryland Medical Center, Baltimore, MD , University of Pittsburgh, Pittsburgh, PA , Department of Neuroradiology, Barrow Institute, Phoenix,AZ , US Naval Academy, Annapolis, MD , Johns Hopkins University, Baltimore, MD , University of Pittsburgh, Pittsburgh, PA , University of Pittsburgh, Pittsburgh, PA , University of Pittsburgh, Pittsburgh, PA , University of Pittsburgh, Pittsburgh, PA , Johns Hopkins University, Baltimore, MD , University of Pittsburgh, Pittsburgh, PA
An Improved Algorithm of White Matter Hyperintensity Detection in Elderly Adults
Automated segmentation of the aging brain poses a significant challenge,to brain tissue classification due to the presence of white matter hyperintensities,(WMH). WMHs appear as hypointense areas in magnetic resonance imaging (MRI),and are frequently found in Alzheimer's disease (AD) population's brain. In recent,years statistical and machine learning methods have been proposed to detect WMH,in the brains of older adults. In this work we proposed an improved version of OASIS,(1) model, an automated statistical inference model for segmentation, initially developed for lesion segmentations in multiple sclerosis (MS) using MRI studies. Our,proposed methods add to the literature in at least two novel ways: 1) by improving the preprocessing stream through the use of an eroding procedure on the skull,stripped mask and 2) by adding a nearest neighbor function and a Gaussian filter to,refine segmentation results. We compared our results with several existing validated,methods based on manual segmentation from experienced radiologists. Using multiple evaluation metrics we found that our improved OASIS model (iOASIS) achieved,competitive or better results compared to other models.
References : 1.Sweeney E et al. (2013). OASIS is Automated Statistical Inference for Segmentation, with applications to multiple sclerosis lesion segmentation in MRI. Neuroimage Clinical, doi.org/10.1016/j.nicl.2013.03.002.
Keywords: Gray matter segmentation, WMH, Alzheimer's, OASIS.
6. Yujia Deng, Annie Qu, Xiwei Tang
University of Illinois, Urbana-Champaign , University of Illinois, Urbana-Champaign , University of Virginia
Correlation Tensor Decomposition and Its Application in Spatial Imaging Data
In this work, we propose a new method to analyze spatial-correlated imaging data. In contrast to the conventional multivariate analysis where the variables are treated as vectors and correlation is represented as a matrix form, we formulate spatial correlation based on the tensor decomposition to preserve the spatial information of imaging data. Specifically, we propose an innovative algorithm to decompose the spatial correlation into a sum of rank-1 tensor such that the structure of the spatial information can be captured more fully compared to traditional approaches. Our method is effective in reducing the dimension of spatial correlated data, which is advantage in computation. In addition, we show that the proposed method can test against the null hypothesis of independent structure, and identifies the block patterns of spatial correlations of imaging data effectively and efficiently. We compare the proposed method with other competing methods through simulations and optical image data to detect early-stage breast cancer.
Keywords: Multidimensional data, Spatial correlation, Tensor decomposition, Dimension reduction, Image processing
7. Hamed Honari, Ann. S. Choe, James J. Pekar, Martin A. Lindquist
Johns Hopkins University, Johns Hopkins University School of Medicine, Johns Hopkins University School of Medicine, Johns Hopkins University.
Investigating the Impact of Autocorrelation on Time-Varying Connectivity
In recent years, a number of studies have reported on the existence of time-varying functional connectivity (TVC) in resting-state functional magnetic resonance imaging (rs-fMRI) data. The sliding-window technique is currently one of the most commonly used methods to estimate TVC. Although previous studies have shown that autocorrelation can negatively impact estimates of static functional connectivity, its impact on TVC estimates is not well known at this time. In this paper, we show both theoretically and empirically that the existence of autocorrelation within a time series can inflate the sampling variability of TVC estimated using the sliding-window technique. This can in turn increase the risk of misinterpreting noise as true TVC and negatively impact subsequent estimation of whole-brain time-varying FC profiles, or "brain states". The latter holds as more variable input measures lead to more variable output measures in the state estimation procedure. Finally, we demonstrate that prewhitening the data prior to analysis can lower the variance of the estimated TVC and improve brain state estimation. These results suggest that careful consideration is required when making inferences on TVC.
Keywords: dynamic functional connectivity, time-varying functional connectivity, resting-state fMRI, autocorrelation, sliding-window, prewhitening
8. Grace Hyun Kim, Yu Shi, Weng Kee Wong, Jonanthan Goldin
Biostatistics and Radiological Science at UCLA
A study design for machine learning technique to predict the likelihood of progression in follow up using baseline chest HRCT images in subjects with idiopathic pulmonary fibrosis
Background : Idiopathic pulmonary fibrosis (IPF) is a fatal lung disease characterized by an unpredictable progressive decline in lung function. A pilot study shows that high resolution computed tomography (HRCT) images can be useful for predicting IPF progression at small regions of interest (ROI) level. We apply the classifier trained at ROI level to the voxels extracted from whole lungs. We hypothesize that a whole lung level metric derived from a baseline scan is associated with progression in a follow-up scan.
Design: We implemented a study design to collect the ground truth of ROI progressive status for supervised learning. Using the paired chest HRCT of baseline and follow-up scan, a radiologist was instructed to contour ROIs at baseline scans and to label ROIs as ‚ expected to be progressed‚ or stable status in follow-up visits. We built a prediction model using machine learning algorithm Quantum Particle Swarm Optimization ‚Äì Random Forest to determine the likelihood of progression in interstitial lung disease as single-scan total probability (STP) in small ROIs using baseline HRCT. Moreover, we extrapolated a STP score in whole lung for each subject.
Methods: We retrospectively collected anonymized longitudinal HRCT images of 193 IPF subjects from multiple clinical trials. Such images were collected for IPF diagnosis purposes. ROIs of 71 subjects were annotated and used to build the classifier, with 82% sensitivity, 65% specificity, and 73% accuracy. HRCT images from 122 subjects were used to test the baseline metric at whole lung level. STP score was tested for association with changes in quantitative lung fibrosis (QLF) scores between two visits, where QLF is a quantification of the disease extent.
Results: The baseline STP is well correlated with the QLF change in the follow-up scans. A univariate analysis shows a normalized hazard ratio of 1.45 (p=0.027). A multivariate analysis controlling subjects‚ age and gender shows a normalized hazard ratio of 1.53 (p=0.041). These results suggest that a higher STP is statistically significantly correlated with higher risk of increased QLF scores in the follow-up scans.
Conclusion : This study design can serve as a reference for collecting ground truth and for developing machine learning techniques to predict progression at follow-up visits.
9. Rejaul Karim , Taps Maiti , Chae Young LIm , Abdhi Sarkar
Michigan State University , Michigan State University , Seoul National University, University of Pennsylvania
Discriminant Analysis for Longitudinal MRI
Linear Discriminant Analysis is one of the popular classification methods. Main goal is sparse representation of Fischer discriminant direction. The asymptotic properties of this classifier is investigated under spatio-temporal paradigm. The results are verified on high dimensional data sets like longitudinal MRI images obtained from ADNI (Alzheimer's Disease Neuroimaging Initiative) archives.
Keywords: High Dimensional , Fischer LDA, Spatio Temporal data,ADNI data, Penalized Likelihhod
10. Dan Kessler, Elizaveta Levina, Keith Levin
University of Michigan
Prediction from networks with node features with application to neuroimaging
While many statistical methods for networks involve single-network tasks (e.g., community detection), here we consider the setting where many networks are observed on a common node set. Each observation is a network, possibly weighted, with covariates at each node, and a network-level response variable. Our motivating application is neuroimaging, where edge weights might represent functional connectivity, node covariates might represent task activation at each location, and network-level response may be disease status or a score on a clinical assessment. The goal is to use the edge weights and node covariates to predict the response and identify a parsimonious and interpretable set of predictive features. We propose a likelihood-based method motivated by the stochastic block model, combined with penalties designed to take into account community structure (naturally occuring in neuroimaging applications) and natural groupings of edges and associated nodes. We provide a theoretical analysis and empirical results on synthetic data demonstrating good performance of the proposed method, and we apply it to data from the human connectome project.
Keywords: Networks, Group Penalty, Brain Imaging, Community Structure, Prediction, Neuroimaging
11. Wenyi Lin
University of California, San Diego
Comparison of Automated Liver Image Quality Evaluation Using Handcrafted Features and Convolutional Neural Networks
Hepatobiliary phase (HBP) MRI using Gd-EOB-DTPA contrast improves detection and characterization of focal hepatic lesions due to increased contrast between liver parenchyma and non-hepatocellular lesions. However, adequate hepatocellular contrast enhancement (HCE) may not be achieved in livers with poor function or when HBP images are acquired early. The purpose of this study was to develop a machine learning approach for classifying HBP images as having adequate or inadequate HCE for detecting focal hepatic lesions.
The data comprised 1201 T1w 3D HBP images from 406 patients who underwent Gd-EOB-DTPA-enhanced liver MRI. Each image was annotated by radiologists as having adequate or inadequate HCE for detecting focal hepatic lesions.
The complete analysis consisted of three parts: image preprocessing to remove bias, handcrafted feature extraction and image classification. Two types of features, intensity separation and topological structure, were captured by Gaussian mixture model and Euler characteristic curves, respectively. 826 HBP images were randomly selected for training and 375 images for validation. Classification was performed using support vector machines and handcrafted performance was compared to a convolutional neural network (CNN). Performance across random subsets of the training dataset were also compared. Performance was evaluated using Receiver operating characteristic (ROC) analysis and Area Under the Curve (AUC).
With complete training data, AUCs were 0.880 (95% CI, 0.84-0.924) for handcrafted features and 0.919 (95% CI, 0.886-0.951) for handcrafted and CNN-learned features. Training on 100 images yielded AUCs of 0.847 (95% CI, 0.805-0.892) for handcrafted features and 0.531 (95% CI, 0.468-0.595) for CNN-learned features. Training on 400 images yielded AUCs of 0.884 (95% CI, 0.847-0.925) for handcrafted features and 0.877 (95% CI, 0.838-0.913) for CNN-learned features.
From the results, handcrafted features outperformed the CNN when training on fewer examples. Increases in training sample size improved both handcrafted and CNN performance. The CNN was more sensitive to changes in training sample size. For large sample sizes, the CNN outperformed the handcrafted features, which is expected.
Keywords: Liver MRI, Quality Evaluation, Machine Learning
12. Melissa Lynne Martin , Russell T. Shinohara
University of Pennsylvania , University of Pennsylvania
Using R to conduct retrospective analyses of EHR and imaging data: A case study in MS
Conducting a retrospective analysis of electronic health records is a challenging task that requires many resources and tools. Increasingly, database resources that are acquired through the clinical care for thousands of patients include longitudinal, research-quality magnetic resonance imaging (MRI). The clinical notes, demographic data, and images obtained from such databases provide an abundance of information for researchers studying neurological diseases. However, these resources come with the informatics challenges of collecting and organizing these retrospective data. As a case study, we led a study based at The Penn Comprehensive Multiple Sclerosis (MS) Center at the University of Pennsylvania to collect over ten thousand MRI studies. We utilized NeuroConductor-based tools in R to organize the information obtained and create an analytic dataset. This is a key step in facilitating the use of these data to answer important questions about MS.
Keywords: R, imaging, multiple sclerosis, informatics, retrospective analysis, electronic health records
13. Fatma Parlak , Amanda F. Mejia
Indiana University , Indiana University
The Role of Experiment, Acquisition Method, Modeling Strategy, and Individual and Spatial Variability on Residual Autocorrelation in Task fMRI Analysis
In a functional resonance imaging (fMRI) experiment, participants‚Äô brains are imaged while they are performing a series of tasks in order to infer brain regions activating in response to each task. To perform this inference, it is typical to fit a linear model at every location relating the observed fMRI data to the expected blood oxygenation level dependent (BOLD) response to each task. However, the assumption of independent residuals in the linear model is violated, resulting in underestimated standard errors. We investigate the factors driving residual autocorrelation, including various experimental factors, individual and spatial differences. We also consider the ability of different modeling strategies to reduce residual autocorrelation by more accurately capturing the shape and duration of the hemodynamic response measured by fMRI, which is known to differ across subjects, areas of the brain and tasks. We find that residual autocorrelation shows conspicuous differences due to experiments and acquisition methods and varies across individuals and areas of the brain. Further, residual autocorrelation is reduced through more flexible HRF modeling approaches.
Keywords: functional resonance imaging, blood oxygenation level dependent, hemodynamic response, linear model
14. Dustin Pluta
University of California, Irvine
Latent Factor Gaussian Process Model for Time-varying Connectivity Analysis of Local Field Potentials
Local field potentials (LFP) measured across a set of tetrodes implanted in the hippocampus of a rat naturally produce connectivity measures that are approximately low-rank. We here present a hierarchical latent variable model for time series of connectivity matrices arising from LFP measurements during a complex cognitive task. By leveraging the low-rank structure of the connectivity data through a latent variable representation, the proposed method efficiently models the time-varying covariance with minimal loss of information. A simulation study verifies performance in scientifically realistic settings. Results of applying our model to LFP data from four rats during a odor-based sequence memory task identify a small set of connectivity features that are most strongly related to the experimental conditions. From these results, we consider an interpretation of our results with respect to the "memory replay" hypothesis of hippocampal function during a sequence memory task.
Keywords: Dynamic connectivity, Bayesian factor model
15. Sarah Ryan
Colorado School of Public Health
The lungct R Package: Software for the Processing and Analysis of Computed Tomography Scans of the Lungs
As part of Neuroconductor, an open-source platform for medical image analysis in R, we have developed the lungct R package. This package focuses on the pre-processing and analysis of computed tomography (CT) scans of the lungs. In this package, we implement a thresholding- and region-based segmentation algorithm for the identification of the left and right lungs from the original scan. Next, using validated functions from ANTsR and other Neuroconductor packages, we develop a pipeline to create disease-specific lung atlases that converge to an unbiased shape and size, as we define by a Dice Similarity Coefficient > 0.99. Using this pipeline, we create a template from a healthy adult population that is publicly available for download. While a variety of options are provided for template creation, we recommend the use of an iterative algorithm using non-linear registration on binary lung segmentations, as these options are flexible enough for the lungs while preserving lung texture. Finally, we provide functions to calculate hundreds of radiomic features on both the two-dimensional slices of CT in the axial, coronal, and sagittal planes, as well as, on three-dimensional regions of interest or the whole-lung. In this project, we demonstrate the full functionality of lungct, by taking chest CTs from raw DICOM to complete statistical analysis.
Keywords: R, open-source, lungs, computed tomography, processing
Mithra T Sathishkumar , Joren Adams , Tallie Z Baram , Michael A Yassa
University of California at Irvine , University of California at Irvine , University of California at Irvine , University of California at Irvine
Brain circuit alterations in the brains of grieving mothers
Rationale Grief is considered an appropriate and time-constrained response to a loss. Yet, mothers who lose their children often experience overwhelming sense of grief that may last for years and decade. The profound grief is often triggered by memories of the lost child or by cues that provoke these memories. Loss of a child upon trauma,war, suicide or drug overdose is common, and the population of grieving mothers is large. Yet, we know little about the neurobiological basis of chronic persistent maternal grief.
Methods Here, we explore the impact of prolonged grief on memory and brain circuitry using structural, functional, diffusion and resting state brain magnetic resonance imaging (MRI).We recruited9 mothers who have experienced a recent loss of a child, (mean age = 61.3 years); and 8 age and demographically matched controls, (mean age = 57.8 years). We assessedcognitive functions, mood and symptoms of grief using. Neuropsychological tests. We employed functional MRI and measured grief and control subjects‚ on brain responses to pictures of their own children (dead for the subjects, alive for the controls), deceased celebrities orunfamiliar people. Subjects also rated the level of emotionality they felt about the images.
Results: Comparing measures of brain activation to their own children vs celebrities or unfamiliar people distinguished grieving mothers from the controls. Specifically,increased activation of components of the prefrontal cortex, and especially both left and right inferior frontal gyrus (IFG) and superior frontal gyrus (SFG) was observed in the grief group.These regions contribute to a network involved in emotional regulation, control, and motivational processing.
Conclusion The disrupted operation of an executive networks balancing motivation (motherhood) and emotional regulationmay begin to uncovera potential basis for the overwhelming and persistent grief of mothers losing their children.
Keywords: fMRI, pre frontal cortex, PFC, grief
17. Daniel Spencer , Rajarshi Gohaniyogi , Raquel Prado
UC Santa Cruz , UC Santa Cruz , UC Santa Cruz
Bayesian Mixed Effect Sparse Tensor Response Regression Model with Joint Estimation of Activation and Connectivity
Brain activation and connectivity analyses in task-based functional magnetic resonance imaging (fMRI) experiments with multiple subjects are currently at the forefront of data-driven neuroscience. In such experiments, interest often lies in understanding activation of brain voxels due to external stimuli and strong association or connectivity between the measurements on a set of pre-specified group of brain voxels, also known as regions of interest (ROI). This poster proposes a joint Bayesian additive mixed modeling framework that simultaneously assesses brain activation and connectivity patterns from multiple subjects. In particular, fMRI measurements from each individual obtained in the form of a multi-dimensional array/tensor at each time are regressed on functions of the stimuli. We impose a low-rank PARAFAC decomposition on the tensor regression coefficients corresponding to the stimuli to achieve parsimony. Multiway stick breaking shrinkage priors are employed to infer activation patterns and associated uncertainties in each voxel. Further, the model introduces region specific random effects which are jointly modeled with a Bayesian Gaussian graphical prior to account for the connectivity among pairs of ROIs. Empirical investigations under various simulation studies demonstrate the effectiveness of the method as a tool to simultaneously assess brain activation and connectivity. The method is then applied to a multi-subject fMRI dataset from a balloon-analog risk-taking experiment in order to make inference about how the brain processes risk.
Keywords: Bayesian inference, brain activation, brain connectivity, functional magnetic resonance imaging, graphical modeling, multiway stick breaking prior, PARAFAC decomposition, tensor response
18. Klaus Telkmann , Babak Shahbaba , Hernando Ombao , Dustin Pluta
University of California Irvine , University of California Irvine , King Abdullah University of Science and Technology , University of California Irvine
Multivariate Functional Data Analysis for LFP Data
We consider samples of multivariate functional data perturbed by white noise. A test for differences in the covariance structure based on functional principal components is proposed. A simulation study and an application to multichannel LFP data illustrate the results.
Keywords: functional data analysis, multivariate, functional principal components
Alessandra Valcarcel , Simon N. Vandekar , Tinashe Tapera , Azeez
Adebimpe , Armin Raznahan , Theodore Satterthwaite , Russell T.
Shinohara , Kristin A. Linn
Perelman School of Medicine, University of Pennsylvania, Vanderbilt,
Approaches for Modeling Spatially Varying Associations Between Multi-Modal Images
Multi-modal magnetic resonance imaging modalities quantify different, yet complimentary, properties of the brain and its activity. When studied jointly, multi-modal imaging data may improve our understanding of the brain. Unfortunately, the vast number of imaging studies evaluate data from each modality separately and do not consider information encoded in the relationships between imaging types. We aim to study the complex relationships between multiple imaging modalities and map how these relationships vary spatially across different anatomical regions of the brain. Given a particular voxel location in the brain, we regress an outcome image modality on the remaining modalities using all voxels in a local neighborhood of the target voxel. In an exploratory analysis, we compare the performance of three estimation frameworks that account for the spatial dependence among voxels in a neighborhood: generalized linear models (GEE), linear mixed effects models with varying random effect structures, and weighted least squares. We apply our framework to a large imaging study of neurodevelopment to study the relationship between local functional connectivity and cerebral blood flow.
Keywords: Multi-Modal, Inter-Modal Coupling
20. Yikai Wang, Ying Guo
Longitudinal Independent Component Modeling Framework for fMRI Decomposition
In recent years, longitudinal neuroimaging study has become increasingly popular in neuroscience research to investigate disease-related changes in brain functions. One important goal in longitudinal imaging analysis is to study temporal changes in brain functional networks (BFN) and its association with subjects' clinical or demographical covariates. In neuroscience literature, one of the most commonly used tools to study BFN is independent component analysis (ICA), which separates multivariate signals into linear mixture of independent components. However, existing ICA methods are not suited for modelling repeatedly measured imaging data. In this paper, we proposed a novel longitudinal independent component model (L-ICA) as the first formal statistical modeling framework that extends ICA to longitudinal setting. By incorporating subject-specific random effects and visit-specific covariate effects, L-ICA is able to provide more accurate estimates of BFNs, borrow information within the same subject to increase statistical power, and allow for model-based prediction. Simulation and real data analysis results demonstrate the advantages of our proposed methods.
21. Ben Wu , Jian Kang , Ying Guo
Emory University , University of Michigan , Emory University
A Thresholded Gaussian Process Independent Component Analysis for Spatially Correlated Sources
Independent component analysis (ICA) which decomposes a multivariate signal into independent non-Gaussian signals has abroad range of applications such as machine learning, finance and neuroimaging. However, for analysis of neuroimaging data, the most existing ICA methods fail to directly account for the spatial dependence among voxels and do not explicitly model the sparsity of source signals. To address those limitations, we propose a Bayesian nonparametric model for ICA of spatial processes. We assume the observed images as the linear mixtures of multiple sparse and spatially correlated latent source processes, for which we construct a new class of prior models by thresholding Gaussian processes. We adopt the von-Mises Fisher distribution as the prior model for mixing coefficients. Under some regularity conditions, we show that the proposed model enjoys large prior support; and we establish the consistency of the posterior distribution with a divergent number of voxels in images. The simulation studies demonstrate that the proposed method outperforms the existing ICA methods for latent brain network separation and brain activation region detection. We apply the proposed method to analysis of the resting-state fMRI data in the Autism Brain Imaging. Data Exchange study and Kirby 21 database.
Keywords: Independent component analysis, Gaussian process, Spatial correlation
22. Yubai Yuan , Annie Qu
University of Illinois, Urbana-Champaign
Community Detection with Dependent Connectivity
In network analysis, it is common that within community is more likely connected than between community, which is reflected by the edges within a community are more correlated. However, the traditional probabilistic models for community detection like stochastic block model (SBM) are not able to capture the dependence among edges. The revised SBM based on random effects can only handle exchangeable dependence structure on whole networks. In this talk, we propose a new community detection approach based on the composite likelihood to utilize the within-community dependence of connectivity. The proposed method allows for the heterogeneity among edges and provides greater flexibility in handling different types of within-community dependence structure. In addition, the proposed algorithm does not involve specifying the likelihood function that could be intractable when correlations exist among edges. We demonstrate the application of the proposed method to the schizophrenia fMRI data. This is joint work with Annie Qu.
Keywords: heterogeneous stochastic block model, Bahadur Representation, high-order approximation, variational EM, product trading network.