Invited talk organized by the Medical Image Analysis lab.
Most current analyses in neuroimaging studies of brain disorders assume a relatively homogenous presentation of the disorder such that traditional statistical analysis methods based on Gaussian distributions can be applied. Yet, almost all brain disorders present with a heterogeneous set or spectrum of cognitive, behavioral, morphometric as well as functional manifestations.
In this talk, we present a novel analysis method for heterogeneous disorders, such as autism spectrum disorder (ASD), via convolutional deep learning based out-of-distribution/anomaly detection. Specifically, we introduce a global and local multi-scale analysis of the gradients of the log density (scores) called Multiscale Score Matching Analysis (MSMA). This method is applied to a structural brain development study of Down Syndrome and ASD to discover and describe a novel heterogeneous set of brain phenotypes via self-organizing maps of the multi-scale anomaly score space.
Dr. Styner is a leading expert in medical image computing with specific research on developing and applying novel neuroimaging methods and software tools for structural and diffusion MRI. His research is mainly focused at studies of early postnatal brain development, encompassing a broad range of studies in human and non-human primate neuroimaging of normal development, autism spectrum disorder, fragile-X, Angelman Syndrome, Down Syndrome, and intra-uterine exposure studies.
Dr. Styner has co-authored over 400 papers in peer-reviewed journals and conferences. He is the director of the UNC Neuro Image Research and Analysis Laboratory (NIRAL).
