Speakers: SeungSuk Kang, PhD
SeungSuk Kang, PhD
Department of Biomedical Sciences
University of Missouri-Kansas City
SeungSuk Kang is an Assistant Professor of the Department of Biomedical Sciences at the University of Missouri-Kansas City (UMKC) School of Medicine. SeungSuk Kang received their BA in Psychology and Philosophy (2000) and MA in Clinical Psychology (2002) from Yonsei University, South Korea, and Ph.D. in Biological Psyhopathology (2009) from the University of Minnesota Twin Cities. Their laboratory focuses on extending our systems-level understanding of the neural mechanisms of aberrant cognitive and affective processes in severe psychopathology, including schizophrenia and post-traumatic stress disorder (PTSD). Ther research also aims to develop alternative interventions for the disorders, including optimized non-invasive neuromodulations and meditation techniques. Their research programs conduct experimental studies with human subjects using electromagnetic neural activity measures, including electroencephalography (EEG) and magnetoencephalography (MEG), brain source imaging of EEG/MEG, and structural and functional MRI data analyses. To solve complex problems of neuroimaging studies, my lab actively uses machine-learning approaches. In particular, their recent research projects utilize deep-learning algorithms to enhance the performance of neuroimaging data processing pipelines and develop novel neuroimaging software algorithms and data-driven clinical diagnosis support systems.
Machine-Learning Approaches in Neuroimaging Studies
Recent explosions of neuroimaging data with advances in technologies have shown the limitations of the conventional theory-driven approach in systems neuroscience and neuroimaging studies. The rapid growth of data-driven and informatics approaches to neuroimaging data, especially machine learning, has demonstrated their utility for efficient data processing and management, the discovery of novel patterns of brain network functions, and the facilitation of developing computational models of the nervous systems and neural processes. In this talk, the unique problems that neuroimaging community is experiencing will be discussed, and three examples of data-driven and machine-learning approaches from my laboratory will be presented. The examples include 1) dimensionality reduction for rigorous EEG and MEG data processing, 2) an image segmentation project for automatic anatomical segmentation of structural MRI scans for studying the human claustrum, the small but the brain’s most highly connected gray matter structure, and 3) the development of computer-aided diagnosis (CAD) system for schizophrenia utilizing 4D neuroimaging of EEG brain source signals. In the end, the caveats in developments of deep-learning algorithms, the advanced form of machine learning for neuroimaging datasets, and the future directions toward neuroinformatics will be discussed.