Dominique Duncan, PhD

Assistant Professor of Neurology, Neuroscience, and Biomedical Engineering
Laboratory of Neuro Imaging
USC Stevens Neuroimaging and Informatics Institute
Keck School of Medicine of USC
University of Southern California

Short Bio:

Dominique Duncan is an assistant professor of Neurology at the USC Stevens Neuroimaging and Informatics Institute in the Laboratory of Neuro Imaging (LONI). Dr. Duncan’s background spans mathematics, engineering, and neuroscience. She double majored in Mathematics and Polish Literature as an undergraduate at the University of Chicago and minored in Computational Neuroscience. She earned her PhD in Electrical Engineering at Yale University. In her PhD thesis, she analyzed intracranial EEG data using nonlinear factor analysis to identify preseizure states of epilepsy patients. After receiving her PhD, she was a professor of Mathematics at Sichuan University in Chengdu, China for a summer program where she taught Calculus 2, Calculus 3, and Linear Algebra to undergraduate students. She then took a postdoctoral position in Neurology at the Stanford University School of Medicine as well as one in Mathematics at UC Davis, where she developed an algorithm based on diffusion maps to classify Alzheimer’s patients using MRI. She has built international, multidisciplinary collaborations and developed novel analytic tools for analyzing multimodal data, including imaging and electrophysiology. Her interests lie at the intersection of data analysis, signal processing, and machine learning, particularly in the areas of traumatic brain injury and epilepsy. By creating large-scale data repositories and linking them with visualization and analytic tools, she aims to encourage collaboration across multiple fields. Dr. Duncan also uses virtual reality to optimize the process of analyzing neuroimaging data and to improve neuroscience education among K-12 students.


The impact of innovative multimodal quantitative methods to control post-traumatic epilepsy


Epileptogenesis is a condition in which an individual develops epileptic seizures and occurs in approximately 15-55% of individuals following a severe traumatic brain injury (TBI), making it among the most widespread disabling disorders of the brain. We have analyzed multimodal imaging and EEG data from the Epilepsy Bioinformatics Study for Antiepileptogenic Therapy (EpiBioS4Rx) from both human patients and an animal model. The goal of this study is to identify biomarkers of epileptogenesis after TBI, then patients can receive antiepileptogenic therapies or treatment to stop or prevent the development of post-traumatic epilepsy (PTE), which would significantly impact the quality of life for TBI patients who are at high risk of developing PTE. Discovering biomarkers of PTE is challenging because the epileptogenic process is multifactorial and requires the integration and analysis of multimodal data, including imaging and EEG.
EpiBioS4Rx will have an enrollment of 300 patients at the end of the study, and the large amount of continuous scalp EEG recordings already collected from more than half of this total number of patients is why we have chosen to apply a manifold learning approach to reduce the number of features while retaining the maximum amount of information about the underlying brain activity of each patient. This allows us to reduce the dimensionality of the data while identifying patterns that can help us distinguish those patients who have developed PTE and those who have not experienced seizures. We have previously developed an algorithm based on a manifold learning approach, Unsupervised Diffusion Component Analysis (UDCA), and we apply this to a subset of the available EEG data. UDCA is adapted to fit the noisy, stochastic nature of scalp EEG data. We also apply various types of filters to automatically-detected high frequency oscillations (HFOs), which may be a biomarker of epileptogenesis. Furthermore, we train 3 machine learning classifiers (Randon Forest, Support Vector Machines, and Neural Networks) on both structural and functional MRI data to distinguish seizure outcome in this TBI patient population. Quantitative and automatic methods are used to extract and investigate multiscale features. These quantitative methods may become useful for clinicians to localize epileptogenic alterations in TBI patients and shed light on the mechanisms that lead to neuronal activity alterations from structural damage.