Advances in Neuroimaging
Presenter: Benjamin Cassidy
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Lecture Theatre (Level 1)
Session Synopsis:
Predicting Parkinson's Disease using individualized whole-brain functional connectivity networks
Non-invasive neuroimaging of brain activity has enormous potential to provide biomarkers for neurological diseases, to identify disease onset before standard clinical symptoms are observed. One approach is to estimate brain networks, and identify abnormal interactions between brain regions, using spatio-temporal functional Magnetic Resonance Imaging (fMRI) data. In this work we demonstrate a combined framework for network inference and disease prediction using resting state fMRI data. We first estimate the link strength between pairs of brain regions, using a new method, DISCOH, to identify whole-brain conditional independence networks from individual fMRI datasets. We then use those link strengths in an Elastic Net penalized (logistic) regression to predict disease status. We show results from a Parkinson's Disease study, demonstrating that the new combined framework can allow better interpretation than using simple time series cross-correlation estimates of the functional connectivity networks, while maintaining the extremely high prediction performance of the correlation based networks. Finally, we use new methods from Topological Data Analysis to also show that the DISCOH network core architectures are more consistent between subjects than correlation networks, despite the high predictive performance for both cases.Non-invasive neuroimaging of brain activity has enormous potential to provide biomarkers for neurological diseases, to identify disease onset before standard clinical symptoms are observed. One approach is to estimate brain networks, and identify abnormal interactions between brain regions, using spatio-temporal functional Magnetic Resonance Imaging (fMRI) data. In this work we demonstrate a combined framework for network inference and disease prediction using resting state fMRI data. We first estimate the link strength between pairs of brain regions, using a new method, DISCOH, to identify whole-brain conditional independence networks from individual fMRI datasets. We then use those link strengths in an Elastic Net penalized (logistic) regression to predict disease status. We show results from a Parkinson's Disease study, demonstrating that the new combined framework can allow better interpretation than using simple time series cross-correlation estimates of the functional connectivity networks, while maintaining the extremely high prediction performance of the correlation based networks. Finally, we use new methods from Topological Data Analysis to also show that the DISCOH network core architectures are more consistent between subjects than correlation networks, despite the high predictive performance for both cases.Non-invasive neuroimaging of brain activity has enormous potential to provide biomarkers for neurological diseases, to identify disease onset before standard clinical symptoms are observed. One approach is to estimate brain networks, and identify abnormal interactions between brain regions, using spatio-temporal functional Magnetic Resonance Imaging (fMRI) data. In this work we demonstrate a combined framework for network inference and disease prediction using resting state fMRI data. We first estimate the link strength between pairs of brain regions, using a new method, DISCOH, to identify whole-brain conditional independence networks from individual fMRI datasets. We then use those link strengths in an Elastic Net penalized (logistic) regression to predict disease status. We show results from a Parkinson's Disease study, demonstrating that the new combined framework can allow better interpretation than using simple time series cross-correlation estimates of the functional connectivity networks, while maintaining the extremely high prediction performance of the correlation based networks. Finally, we use new methods from Topological Data Analysis to also show that the DISCOH network core architectures are more consistent between subjects than correlation networks, despite the high predictive performance for both cases.Non-invasive neuroimaging of brain activity has enormous potential to provide biomarkers for neurological diseases, to identify disease onset before standard clinical symptoms are observed. One approach is to estimate brain networks, and identify abnormal interactions between brain regions, using spatio-temporal functional Magnetic Resonance Imaging (fMRI) data. In this work we demonstrate a combined framework for network inference and disease prediction using resting state fMRI data. We first estimate the link strength between pairs of brain regions, using a new method, DISCOH, to identify whole-brain conditional independence networks from individual fMRI datasets. We then use those link strengths in an Elastic Net penalized (logistic) regression to predict disease status. We show results from a Parkinson's Disease study, demonstrating that the new combined framework can allow better interpretation than using simple time series cross-correlation estimates of the functional connectivity networks, while maintaining the extremely high prediction performance of the correlation based networks. Finally, we use new methods from Topological Data Analysis to also show that the DISCOH network core architectures are more consistent between subjects than correlation networks, despite the high predictive performance for both cases.
Advances in Neuroimaging
Presenter: John Hughes
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Lecture Theatre (Level 1)
Session Synopsis:
Fast, fully Bayesian spatiotemporal inference for fMRI data
We propose a spatial Bayesian variable selection method for detecting blood oxygenation level dependent (BOLD) activation in functional magnetic resonance imaging (fMRI) data. Typical fMRI experiments generate large datasets that exhibit complex spatial and temporal dependence. Fitting a full statistical model to such data can be so computationally burdensome that many practitioners resort to fitting oversimplified models, which can lead to lower quality inference. We develop a full statistical model that permits efficient computation. Our approach eases the computational burden in two ways. We partition the brain into three-dimensional parcels, and fit our model to the parcels in parallel. Voxel-level activation within each parcel is modeled as regressions located on a lattice. Regressors represent the magnitude of change in blood oxygenation in response to a stimulus, while a latent indicator for each regressor represents whether the change is zero or nonzero. A sparse spatial generalized linear mixed model (SGLMM) captures the spatial dependence among indicator variables within a parcel and for a given stimulus. The sparse SGLMM permits considerably more efficient computation than does the spatial model typically employed in fMRI. Through simulation we show that our parcellation scheme performs well in various realistic scenarios. Importantly, indicator variables on the boundary between parcels do not exhibit edge effects. We conclude by applying our methodology to data from a task-based fMRI experiment.
Advances in Neuroimaging
Presenter: Galin Jones
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Lecture Theatre (Level 1)
Session Synopsis:
Estimating Neuron Fiber Orientation via Diffusion MRI
We present a framework for estimating neuron fiber orientations with diffusion tensor imaging (DTI). A commonly used model in DTI is a simple partial volume model of diffusion using a Bayesian model whose parameters define local neuron fiber direction. We will show that this popular model results in an improper posterior distribution so that no valid inference may be performed. We also show how to modify the model so that the posterior will be proper. Markov Chain Monte Carlo (MCMC) is implemented to fit the model and rigorous methods are developed for ascertaining how many Monte Carlo samples are required to ensure reliable estimation. We apply our method to an example and show that the Monte Carlo sample sizes typically used in applications are too small.
Advances in Neuroimaging
Presenter: Ani Eloyan
When: Thursday, July 14, 2016 Time: 4:30 PM - 6:00 PM
Room: Lecture Theatre (Level 1)
Session Synopsis:
Statistical issues in pre-processing and group modeling of magnetic resonance image data for neurodegenerative diseases
Statistical analysis of medical imaging data including magnetic resonance imaging (MRI), computed tomography (CT), etc. highly depends on the quality of the collected data. Even though each imaging modality presents its own set of data contamination sources, there are general issues that appear consistently in most studies. These include the effects of movement in the scanner, systematic noise, alignment of images to a common template and normalization of intensities to a common scale. In this talk, some of the issues in statistical analysis of high dimensional medical imaging data are discussed, along with methods for improving the statistical analysis for disease exploration.
