Deep learning of Kawasaki Disease outbreaks using dust and meteorological parameters as possible predictors

Abstract Winner: Cloud Computing & Big Data

Author: Hesham El-askary

Kawasaki disease (KD) outbreaks resulting in permanent heart damage, puzzled a great deal of medical doctors and recently scientists on the possible causes of such outbreaks. KD is characterized by seasonal outbreaks that matches other seasonality associated with Earths systems processes that can, if not at least induce, be a vital player in the out speared. A fungus known as Candida originating in the farmlands of China, that is long range transported with aerosols by different wind mechanisms, has been blamed for it. Over the last decade aerosols have been studied on different scales either regionally or globally using satellite observations owed to their impact on human health and other different sectors. We have used different aerosol related parameters, namely angstrom exponent and fine mode fraction (effective radius 0.1–0.25 microns), to study the anomalous behavior of dust particles associated with KD outbreaks in Japan. Dust particles that have the ability to transport long distances, have a mid-range Angstrom Exponent in a range of [0, 4]. Here we are shedding the light on how we can use machine learning tools namely, Neural Networks and deep learning techniques on earth process driven data sets to early forecast possible KD outbreaks. Traditionally, applications which leverage neural networks as their underlying predictive models have been constrained to shallow network topologies, often consisting of only a single hidden layer. While neural networks with a single hidden layer have been mathematically proven to be universal approximators capable of learning any continuous function, a single hidden layer is often inadequate to learn high order features embedded in imagery or high-dimensional temporal data.

Co Author/Co-Investigator Names/Professional Title: Hesham El-Askary1,2,3*, Nick LaHaye1,4, Erik Linstead1, Magdi Yacoub5 1Schmid College of Science and Technology, Chapman University, Orange, CA, USA 2Center of Excellence in Earth Systems Modeling & Observations, Chapman University, Orange, CA, USA 3Department of Environmental Sciences, Faculty of Science, Alexandria University, Moharem Bek, Alexandria, Egypt 4Jet Propulsion Laboratory, NASA, CA, USA 5 Faculty of Medicine, National Heart & Lung Institute, Imperial College of London