Title: Inference of Structural Brain Networks and Modeling of Cortical Multi-Sensory Integration
School of Computer Science
Georgia Institute of Technology
Date: Thursday, November 15, 2018
Time: 9:30am-11:30am (EST)
Location: Klaus 3402
Prof. Constantine Dovrolis (Advisor), School of Computer Science, Georgia Institute of Technology
Prof. Ellen W. Zegura, School of Computer Science, Georgia Institute of Technology
Prof. Shella Keilholz, Department of Biomedical Engineering, Georgia Institute of Technology and Emory University
Prof. Zsolt Kira, School of Interactive Computing, Georgia Institute of Technology
Prof. Eva L. Dyer, Department of Biomedical Engineering, Georgia Institute of Technology and Emory University
Recent advances in neuroimaging have enabled major progress in neuroscience, especially the field of brain connectomics, i.e., comprehensive maps of connections between brain regions at different scales. Diffusion MRI (dMRI) and probabilistic tractography algorithms are state of the art methods to map the structural connectome of the brain non-invasively and in vivo. Although probabilistic tractography has been shown to detect many major connections in the brain, it also reports many spurious ones. We propose and evaluate a method, referred to as MANIA (Minimum Asymmetry Network Inference Algorithm) that can infer the structural brain network that interconnects a given set of Regions of Interest (ROIs) from probabilistic tractography data. Given that diffusion MRI is unable to detect the direction of each connection, we formulate the network inference process as an optimization problem that minimizes the (appropriately normalized) asymmetry of the observed network.
This Ph.D. Proposal outlines two additional research problems we propose to investigate:
The first problem relates to structural brain network inference, and in particular, to the well known "distance bias” that tractography methods introduce. Specifically, the application of tractography methods on dMRI data makes it harder to discover connections between distal regions compared to proximal regions. We propose a machine learning method that models and removes this distance bias from tractography data. We apply the proposed method in tandem with MANIA on data collected by the Human Connectome Project (HCP). Interestingly, we find that the meso-scale human connectome is much denser than previously reported, after the tractography distance bias is removed.
Having a structural network representation is instrumental in analyzing communication dynamics and information processing in the brain. The second research problem we are working on relates to multi-sensory integration in the cortex. We model this process on the mouse meso-scale connectome (mapped by the Allen Institute) applying an Asynchronous Linear Threshold (ALT) diffusion model on that connectome. The ALT model captures how evoked activity that originates at a primary sensory region of the cortex “ripples through” other cortical regions. Our preliminary results show that a small number of brain regions (including the Claustrum) integrate all sensory information streams, suggesting that the cortex relies on an hourglass architecture to integrate and compress multi-sensory information.