Abstract

Deep neural networks (DNN) have recently gained a lot of interest because of their ability to learn representations and high-level abstractions directly from data. Their use is particularly applicable in biomedical signal processing because of the characteristic structure of the corresponding signals. We use an autoassociative DNN to model the Electrodermal Activity (EDA), a physiological signal that depicts typical shape fluctuations over time and responds to the sympathetic nervous system (SNS) activation. The analysis of EDA signals was motivated by the sensory difficulties of children with Autism Spectrum Disorder during dental cleaning by monitoring physiological stress and anxiety in two dental environments. Compared to traditionallyused skin conductance level and response measures, EDA features automatically derived by DNN capture signal patterns in a more reliable way. These results provide a foundation towards accurate automaticallyderived measures of stress and anxiety fro...

Date

2016

Authors

Leah I Stein Duker