Thursday, Aug 8: 8:30 AM - 10:20 AM
5186
Contributed Papers
Oregon Convention Center
Room: CC-B119
Main Sponsor
Section on Statistics in Imaging
Presentations
In fMRI, capturing brain activity during a physical task is dependent on how quickly each volume k-space array is obtained. Acquiring the full k-space arrays can take a considerable amount of time. Under-sampling k-space reduces the acquisition time, but results in aliased, or "folded", images after applying the inverse Fourier transform (IFT). GeneRalized Autocalibrating Partial Parallel Acquisition (GRAPPA) and SENSitivity Encoding (SENSE) are parallel imaging techniques that yield full images from subsampled arrays of k-space. With GRAPPA operating in the spatial frequency domain and SENSE in image space, these techniques can be fused to reconstruct the subsampled k-space arrays more accurately. Here, we propose a Bayesian approach to this combined model where prior distributions for the unknown parameters are assessed from a priori k-space arrays. The prior information is utilized to estimate the missing spatial frequency values, unalias the voxel values from the posterior distribution, and reconstruct into full field-of-view images. Our Bayesian technique successfully reconstructed a simulated fMRI time series with no aliasing artifacts while decreasing temporal variation.
Keywords
Bayesian
GRAPPA
fMRI
reconstruction
SENSE
Functional connectivity, the study of coordination between functionally distinct brain regions, is a recent focus in neuroscientific research. The Psychophysiological Interaction (PPI) model is commonly used to infer functional connectivity in a task-dependent context, but its main limitation is its susceptibility to confounding effects. We argue that partial correlations, rather than the regression coefficients of the PPI model, are a better measure of functional connectivity because they correct for confounding. We show that the PPI model entails a Gaussian Graphical Model (GGM) from which partial correlations are easily derived. We fit our model efficiently with a set of independent Bayesian linear regressions performed in parallel. We allow the regression coefficients to vary over time which accommodates dynamic background connectivity, overcoming another limitation of the PPI model. A thoughtful choice of scale-mixture shrinkage priors enforces sparsity in the GGM precision matrix and discourages overfitting. We demonstrate the efficacy of our model over the PPI model using simulated data and apply it to human fMRI data from a serial reaction time experiment.
Keywords
Phsychophysiological Interaction
Bayesian
Time-varying coefficient
fMRI
Bayesian methods are commonly applied to solve image analysis problems such as noise-reduction, feature enhancement and object detection. A limitation of these approaches is the computational complexity due to the interdependence of neighboring pixels which limits the efficiency of performing full posterior sampling through Markov chain Monte Carlo (MCMC). To alleviate this, we develop a new posterior sampling method that is based on modeling the prior and likelihood in the space of the Fourier transform of the image. One advantage of Fourier-based methods is that a large set of spatially correlated processes in image space can be represented via independent processes over Fourier space. A recent approach known as Bayesian Image Analysis in Fourier Space (or BIFS), has introduced the concept of parameter functions to describe prior expectations about distributional parameters over Fourier space. The work presented here extends BIFS to a posterior sampling approach that can explore a range of posterior estimators beyond the MAP estimate. Computational efficiency of MCMC for BIFS is much improved over that for conventional Bayesian image analysis and mixing concerns are avoided.
Keywords
Bayesian image analysis
Markov chain Monte Carlo
Multi-subject models of the tumor microenvironment (TME) typically rely on global features of imaged tissues and model TME dynamics ignoring specific local spatial changes in cell interactions during tumor development. We introduce a novel method of Bayesian Inference for Point Patterns in Space (BIPPS) for interpretable analyses of the complex spatial dynamics of the TME with multi-subject, multi-tissue multiplexed immunofluorescence (mIF) data. Unlike scope limited black-box methods requiring significant computational resources and posing interpretability issues, BIPPS can model the spatial dynamics of the TME using localized tissue features in an interpretable log-linear spatial factor model for multivariate counts. Each tissue image is expressed as a linear combination of a few subject-specific scalable Gaussian Processes. Covariates characterizing the TME can be integrated into the factor loadings, facilitating efficient estimation of spatial intensities across subjects. We demonstrate BIPPS on an internal cohort of mIF images from patients belonging to six different pancreatic diseases. An open source R package for implementing BIPPS will be available on github.
Keywords
Bayesian
Spatial
Point Pattern
Multi-subject
Imaging
We consider the problem of analyzing multivariate time series collected on multiple subjects, with the goal of identifying groups of subjects exhibiting similar trends in their recorded measurements over time as well as time-varying groups of associated measurements. We propose a Bayesian model for temporal bi-clustering featuring nested partitions, where a time-invariant partition of subjects induces a time-varying partition of measurements. Our approach allows for data-driven determination of the number of subject and measurement clusters as well as estimation of the number and location of changepoints in measurement partitions. To efficiently perform model fitting and posterior estimation with Markov Chain Monte Carlo, we derive a blocked update of measurements' cluster-assignment sequences.
We illustrate the performance of our model in two applications to functional magnetic resonance imaging data and to an electroencephalogram (EEG) dataset. The results indicate that the proposed model can combine information from potentially many subjects to discover a set of interpretable, dynamic patterns.
Keywords
Bayesian
Time Series
Neuroimaging
Clustering
Co-Author(s)
Jaylen Lee, University of California, Irvine
Michele Guindani, University of California-Los Angeles
Marina Vannucci, Rice University
Megan A.K Peters, University of California, Irvine
Sana Hussain, University of California, Riverside
Erik Sudderth, University of California, Berkeley
First Author
Federica Zoe Ricci, University of California Irvine
Presenting Author
Jaylen Lee, University of California, Irvine
Current literature on joint models can typically jointly analyze one or a few longitudinal processes and a time-to-event outcome. We develop a Bayesian semiparametric functional joint model that(1)models high-dimensional longitudinal processes with identifying trajectory based on latent classes nested within each process,(2)provides flexibility in modeling the association between the longitudinal processes and time-to-event outcome,and(3)addresses selection from the high-dimensional longitudinally processes in a global-local way where processes are selected globally and a local selection is used to select the effects of latent classes within each process. This work is motivated by high-dimensional imaging features of the eye, measured longitudinally at multiple visits of patients with early-stage age-related macular degeneration(AMD). A primary scientific question is selection of longitudinal feature processes that can prognosticate conversion to neovascular AMD. Our simultaneous analysis of all imaging features in the proposed model highlights unique features associated in multiple ways with prognostication of conversion to neovascular AMD that are distinct from previous findings.
Keywords
Joint Modeling
High-dimensional
Functional modeling
Bayesian Non-parametrics
Age related Macular degeneration (AMD)
Ophthalmology