Speaker:
Martin Frasch, MD, PhD, is an experienced principal investigator. He is a Research Affiliate at the Center on Human Development and Disability and an Affiliate Assistant Professor at the Department of Obstetrics and Gynecology at the University of Washington, Seattle, USA, and a Visiting Professor at the Technical University of Munich, Germany. He is also an entrepreneur with a focus on Digital Health. Martin’s research has been on the exciting intersection of physiology, health biometrics, wearables, maternal, child and seniors’ health, biosignal processing and health outcomes prediction: from bench to bed, incl. product development and commercialization.
His work on physiological monitoring has been internationally recognized with over 200 publications, memberships in international research societies, service on the editorial boards of multiple top-tier academic journals and research grant review panels worldwide.
Visit his website and LinkedIn for more information: FraschLab.org & www.linkedin.com/in/mfrasch
Abstract:
I will begin the conversation by positioning machine learning (ML) and deep learning (DL) in the field of preclinical and clinical physiological modeling of health outcomes, especially early detection of brain injury. I will present examples of top-down and bottom-up approaches using DL to detect fetal distress as well as of featurization of physiological time series in ML frameworks using all five signal-analytical domains of variability to identify physiological phenotypes. I will place it next to an example of two symbolic logic approaches to computational systems physiology, exemplified by an in silico model of labor and coordination dynamics/metabolic optimization. Finally, closing the conversation I will loop from developmental neuroscience back to ML discussing the implications of a fascinating deep connection between developmental neuroscience and neural architecture search (NAS). Where does this leave us with regard to physiological causal reasoning from “black box” deep learning? Can we open that box and offer clinicians interpretable real-time decision support for managing pregnant patients? Last but not least, can insights from developmental neuroscience offer a new perspective on NAS and the foundational assumptions in ML modeling more generally?
