SPS BISP TC Webinar: Brain Hacking Rare Vascular Disease: Novel Methods for Brain Age, White Matter Hyperintensity and Functional Connectivity on a Newly Collected CADASIL Dataset

Date: 14 July 2023
Time: 10:00 AM EDT (Local time)
Presenter(s): Dr. Marlena Duda, Bradley T. Baker


Vascular cognitive decline can manifest in a diverse array of symptoms observable in brain imaging data sets. In this work, we present the results of a hackathon aimed at applying several machine-learning methods to identify and track biomarkers within a new consortium of subjects exhibiting Cerebral Autosomal Dominant Arteriopathy with Subcortical Infarcts and Leukoencephalopathy (CADASIL), an inheritable cerebrovascular disease which affects arterial walls and can manifest in white matter lesions detectable with magnetic resonance imaging (MRI).
We first performed group difference analysis of several derivatives from structural, functional, and diffusion MRI gathered from the CADASIL consortium. In our analysis we found significant group differences between CADASIL and control groups in White Matter Hyperintensity (WMH) lesion load, dynamic and static FNC states, and peak mean diffusivity metrics.

We also demonstrate a new approach to brain age (BA) prediction using dynamic functional network connectivity (dFNC) data. Our method utilizes a Bi-Directional Long-Short Term Memory (Bi-LSTM) deep neural network trained on dFNC derivatives of the UKBiobank and Human Connectome Project (HCP) datasets. Applying the trained model to holdout data reveals significant correspondence between BA delta (predicted - chronological BA) and clinical scores related to cognitive decline. In contrast to prevailing methods in BA prediction that utilize structural imaging biomarkers, our results suggest that connectivity dynamics may also play an important role in BA prediction. Our method can potentially provide a new perspective on the functional trajectory of aging, which can be useful across a variety of age-related applications, including CADASIL progression.


Anubha GuptaMarlena Duda Marlena Duda received a B.S. degree in biology at Northeastern University, Boston, MA, USA in 2013. She then completed a concurrent M.A. in statistics and Ph.D. in bioinformatics at University of Michigan, Ann Arbor, MI, USA in 2021. She is currently a Postdoctoral Researcher at the TReNDS Center of Georgia State University, Georgia Institute of Technology, Emory University, Atlanta, GA, USA. Broadly, her research interests involve developing algorithms to aid in the detection and understanding of psychiatric disorders, as well as functional brain dynamics in general. Her current research efforts center on the development of multimodal fusion models for the integration of various neuroimaging modalities with additional datatypes, including genetic and microbiome data.



Anubha GuptaBradley Baker Bradley T. Baker received his B.A. degree in Philosophy/Mathematics from the New College of Florida, Sarasota, Florida, in 2016. He received his M.S. degree in Computer Science from University of New Mexico, Albuquerque, New Mexico in 2019. He is currently pursuing the Ph.D. degree in Computational Science and Engineering at Georgia Institute of Technology, Atlanta, GA, USA. His research interests include leveraging insights from optimization theory and neural computation to interpret and innovate on Artificial Neural Networks, distributed learning in privacy sensitive settings for neuroimaging, and developing flexible and powerful software toolboxes for neuroimaging research.