SMI 2022 - Student paper award winners

Wednesday, May 25, 2022 • 4:15–5:30 pm (CT) • Light Hall, Room 202

Chair: Simon Vandekar, Vanderbilt University


Heterogeneity analysis on multi-state brain functional connectivity and adolescent neurocognition [winner]

Shiying Wang, Yale University

with Todd Constable, Heping Zhang, and Yize Zhao*

Brain functional connectivity or connectome, a unique measure for brain functional organization, provides a great potential to reveal the neurobiological underpinning of behavioral profiles. Existing connectome-based analyses highly concentrate on brain activities under a single cognitive state, and fail to consider heterogeneity when attempting to characterize brain-to-behavior relationships. In this work, motivated by a recent landmark brain development and child health study, we study the complex impact of multi-state functional connectivity on behaviors. By proposing a Bayesian supervised heterogeneity analysis, we uncover the neurobiological subtypes nonparametrically and impose stochastic block structures to identify network-based functional phenotypes. A variational expectation–maximization algorithm is developed to facilitate an efficient posterior computation. Through integrating resting and task-related functional connectomes, our data analyses dissect heterogeneous predictive mechanisms on children’s fluid intelligence from the functional network phenotypes, including fronto-parietal network and default mode network, under different cognitive states. Meanwhile, we also achieve a dramatic improvement on prediction using our method compared with existing alternatives and single state analyses. Based on extensive simulations, we further confirm the superior performance of our method on uncovering and predicting brain-to-behavior relationships.


Multi-task learning with high-dimensional noisy images [winner]

Xin Ma, Emory University

with Suprateek Kundu*

Recent medical imaging studies have given rise to distinct but inter-related datasets corresponding to multiple experimental tasks or longitudinal visits. Standard scalar-on-image regression models that fit each dataset separately are not equipped to leverage information across inter-related images, and existing multi-task learning approaches are compromised by the inability to account for the noise that is often observed in images. We propose a novel joint scalar-on-image regression framework involving wavelet-based image representations with grouped penalties that are designed to pool information across inter-related images for joint learning, and which explicitly accounts for noise in high-dimensional images via a projection-based approach. In the presence of non-convexity arising due to noisy images, we derive non-asymptotic error bounds under non-convex as well as convex grouped penalties, even when the number of voxels increases exponentially with sample size. A projected gradient descent algorithm is used for computation, which is shown to approximate the optimal solution via well-defined non-asymptotic optimization error bounds under noisy images. Extensive simulations and application to a motivating longitudinal Alzheimer's disease study illustrate significantly improved predictive ability and greater power to detect true signals, that are simply missed by existing methods without noise correction due to the attenuation to null phenomenon.


Bayesian image-on-scalar regression with a spatial global-local spike-and-slab prior [runner-up]

Zijian Zeng, Rice University

with Meng Li and Marina Vannucci*

In this article, we propose a novel spatial global-local spike-and-slab selection prior for image-on-scalar regression. We consider a Bayesian hierarchical Gaussian process model for image smoothing, which uses a flexible Inverse-Wishart process prior to handle within-image dependency, and propose a general global-local spatial selection prior that extends a rich class of well-studied selection priors. Unlike existing constructions, we achieve simultaneous global (i.e, at covariate-level) and local (i.e., at pixel/voxel-level) selection by introducing participation rate parameters that measure the probability for the individual covariates to affect the observed images. This along with a hard-thresholding strategy leads to dependency between selections at the two levels, introduces extra sparsity at the local level, and allows the global selection to be informed by the local selection, all in a model-based manner. We design an efficient Gibbs sampler that allows inference for large image data. We show on simulated data that parameters are interpretable and lead to efficient selection. Finally, we demonstrate performance of the proposed model by using data from the Autism Brain Imaging Data Exchange (ABIDE) study (Di Martino et al., 2014). To our knowledge, the proposed model construction is the first in the Bayesian literature to simultaneously achieve image smoothing, parameter estimation and a two-level variable selection for image-on-scalar regression.


A structured multivariate approach for removal of batch effects [runner-up]

Rongqian Zhang, University of Toronto

with Lindsay D. Oliver, Aristotle N. Voineskos, and Jun Young Park*

Combining data collected from multiple studies is becoming common and is advantageous to researchers to increase the reproducibility of scientifc discoveries. However, at the same time, unwanted batch effects are commonly observed across neuroimaging data collected from multiplestudy sites or scanners, rendering difficulties in combing such data to obtain reliable findings. While methods for handling such unwanted variations have been proposed recently, most of them use univariate approaches which would be too simple to capture all sources of batch effects which could be represented by the batch-specific latent patterns. In this paper, we propose a novel multivariate harmonization method, called UNIFAC harmonization, for estimating and removing both explicit and latent batch effects. Our approach is based on the simultaneous dimension reduction and factorization of interlinked matrices through a penalized objective, which provides a new direction in neuroimaging research for harmonizing multivariate features across batches. Using the Social Processes Initiative in Neurobiology of the Schizophrenia (SPINS) dataset and extensive simulation studies, we show that UNIFAC harmonization performed better than the existing methods in entirely removing batch effects as well as retaining associations of interest to increase statistical power.


Tensor-variate elliptically contoured distributions with application to image learning [runner-up]

Carlos Llosa-Vite, Iowa State University

with Ranjan Maitra*

Statistical analysis of tensor-valued data has largely involved the tensor-variate normal (TVN) distribution that may be inadequate for modeling data from distributions with heavier or lighter tails. We study a general family of elliptically contoured (EC) random tensor distributions that generalize the TVN distribution. We derive properties of the tensor-variate EC family such as characterizations, marginal and conditional distributions, moments, and the EC Wishart distribution. Next, we describe procedures for maximum likelihood (ML) estimation of parameters from data that are (1) uncorrelated draws from an EC distribution, (2) from a scale mixture of the TVN distribution, and (3) from an underlying but unknown EC distribution, where we extend Tyler’s robust estimator to the tensor-variate setting. A detailed simulation study highlights the benefits of choosing an EC distribution over the TVN distribution for heavier-tailed tensor-variate data. We develop classification rules using discriminant analysis and EC errors and show that they have better predictive performance over those using TVN errors in identify cats and dogs from images in the Animal Faces-HQ dataset. Further, a reduced-rank tensor-on-tensor regression and tensor-variate anallysis of variance (TANOVA) framework under EC errors is demonstrated to better characterize gender, age and ethnic origin than the usual TVN-based TANOVA in the celebrated Labeled Faces of the Wild dataset.


* Asterisks denote primary mentors


Return to the schedule page