Sleep Classification with Artificial Synthetic Imaging Data Using Convolutional Neural Networks

• Activity Number: 307 - Deep Learning, Nonparametric Statistics, and Beyond
• Type: Invited
• Date/Time: Tuesday, August 9, 2022 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Peter Song* and Lan Shi and Marianthie Wank and Yan Chen and Yibo Wang and Emily Charlotte Hector
• Companies: University of Michigan and University of Michigan and University of Michigan and University of Michigan and University of Michigan and North Carolina State University
• Keywords: Classification ; CNN; Deep learning; Sleep health ; Wearable device

Abstract

In sleep studies, the most fundamental question is the detection of sleep-awake status. Wearable devices with multi-modal sensors enable real-time health monitoring and offer a convenient way to track sleep. We propose a new analytic framework, termed as ASID Workflow, for sleep classification consisting of: 1) the creation of artificial synthetic imaging data (ASID) from temporal data collected by E4 wristband that permits monitoring of heart rate (HR), accelerometer, electrodermal activity, and skin temperature, denoted as Temporal E4 Data (TED) and 2) the use of convolutional neural network (CNN) to classify periods of sleep. Competing machine learning algorithms, including logistic regression, support vector machine, and random forest, are applied to TED as comparisons to highlight the power of the ASID Workflow. We also study the influence of data resolution and HR modality on classification accuracy. The results show that the ASID Workflow is superior to the Competing Workflow on data with high resolution and without HR modality, whose classifiers achieve excellent performances with 95% accuracy.