**Note: This is a virtual presentation. Here is the link for where the presentation will be taking pl
ace.**

**Title: **Mapping Brain Connectivity t
o Behavior: from Network Optimization Frameworks to Deep-Generative Hybrid
Models

**Abstract: **Autism Spectrum Disorder (ASD)
is a complex neurodevelopmental disorder characterized by multiple impairm
ents and levels of disability that vary widely across the ASD spectrum. Cu
rrently\, the most common methods of quantifying symptom severity are almo
st solely based on a trained clinician’s evaluation. Recently\, neuroimagi
ng techniques such as resting state functional MRI (rs-fMRI) and Diffusion
Tensor Imaging (DTI) have been gaining popularity for studying aberrant b
rain connectivity in ASD. My thesis aims at linking the symptomatic charac
terization of ASD with the functional and structural organization of a typ
ical patient’s brain as given by rs-fMRI and DTI respectively. My talk is
organised into two main parts\, as follows:

**Network Optimi
zation Models for rs-fMRI connectomics and clinical severity:**

\nAnalysis of a multi-subject rs-fMRI imaging study often begins at the
group level\, for example\, estimating group-averaged functional connectiv
ity across all subjects. The failure of data-driven machine learning techn
iques such as PCA\, k-PCA\, SVMs etc. are largely attributed to their fail
ure at capturing both the group structure and the individual patient varia
bility\, due to which they fail to generalize to unseen patients. To overc
ome these limitations\, we developed a matrix factorization technique to r
epresent the rs-fMRI correlation matrices by decomposing them into a spars
e set of representative subnetworks modeled by rank one outer products. Th
e subnetworks are combined using patient-specific non-negative coefficient
s. The network representations are fixed across the entire group\, however
\, the strength of the subnetworks can vary across individuals. We signifi
cantly extend prior work in the area by using these very network coefficie
nts to simultaneously predict behavioral measures via techniques ranging f
rom simple linear regression models to parametric kernel methods\, to Arti
ficial Neural Networks (ANNs). The main novelty of the algorithms lies in
jointly optimizing for the regression/ANN weights in conjunction with the
rs-fMRI matrix factors. By leveraging techniques from convex and non-conve
x optimization\, these frameworks significantly outperform several state-o
f-the art machine learning\, graph theoretic and deep learning baselines a
t generalization to unseen patients.

**Deep-Generative Hybri
d Frameworks for Integrating Multimodal and Dynamic Connectivity with Beha
vior:**

\nThere is now growing evidence that functional connec
tivity between regions is a dynamically process evolving over a static ana
tomical connectivity profile\, and that modeling this evolution is crucial
to understanding ASD. Thus\, we propose an integrated deep-generative fra
mework\, that jointly models complementary information from resting-state
functional MRI (rs-fMRI) connectivity and diffusion tensor imaging (DTI) t
ractography to extract predictive biomarkers of a disease. The generative
part of our framework is a structurally-regularized Dynamic Dictionary Lea
rning (sr-DDL) model that decomposes the dynamic rs-fMRI correlation matri
ces into a collection of shared basis networks and time varying patient-sp
ecific loadings. This matrix factorization is guided by the DTI tractograp
hy matrices to learn anatomically informed connectivity profiles. The deep
part of our framework is an LSTM-ANN block\, which models the temporal ev
olution of the patient sr-DDL loadings to predict multidimensional clinica
l severity. Once again\, our coupled optimization procedure collectively e
stimates the basis networks\, the patient-specific dynamic loadings\, and
the neural network weights. Our hybrid model outperforms state-of-the-art
baselines in a cross validated setting and extracts interpretable multimod
al neural signatures of brain dysfunction in ASD.

In recent years\ , graph neural networks have shown great promise in brain connectivity res earch due to their ability to underscore subtle interactions between commu nicating brain regions while exploiting the underlying hierarchy of brain organization. To conclude\, I will present some ongoing explorations based on end-to-end graph convolutional networks that directly model the evolut ion of the rs-fMRI signals/connectivity patterns over the underlying anato mical DTI graphs.

\n**Committee Members**

Archa na Venkataraman\, Department of Electrical and Computer Engineering

\n< p>Rene Vidal\, Department of Biomedical Engineering\nCarey E. Prieb e\, Department of Applied Mathematics & Statistics

\nStewart Mostofs ky\, Director of Center for Neurodevelopmental and Imaging Research\, Kenn edy Krieger Institute

\nKilian Pohl\, Program Director\, Image Analy sis\, Center for Health Sciences\,and Biomedical Computing\, SRI Internati onal\; Associate Professor of Psychiatry and Behavioral Sciences\, Stanfor d University

\n\n X-INSTANT-EVENT:1 END:VEVENT END:VCALENDAR