PFI-TT: A New Deep Neural Network and Internet of Things-based Screening Framework for Sleep Apnea

点击翻译按钮获取中文摘要

The broader impact/commercial potential of this Partnerships for Innovation - Technology Translation (PFI-TT) project is to improve sleep apnea screening precision. More than 22 million Americans are affected by sleep apnea. The current home sleep test systems tend to underestimate sleep apnea severity with a high false negative rate. Eighty percent of patients with moderate to severe obstructive sleep apnea are never diagnosed. Because sleep apnea is a serious medical condition and a major public health issue, we urgently need a new sleep apnea screening framework. The proposed framework is one of the first deep learning based sleep apnea screening framework using wearable and IoT sensors. It has commercial potential: it can potentially improve screening accuracy by 40 percent with low cost compared with existing sleep apnea screening devices. The proposed project also includes a broader participant plan. In order to further enhance inclusion of students from all backgrounds, the PFI-TT project will leverage existing programs with schools and programs at the University of Arizona. Additional workshops and teleconferences will be provided for underrepresented groups, particularly women, Hispanic, African Americans and Native Americans, to mentor and coach students with interests in entrepreneurship. The success of our program can be adapted at other institutions nationally and globally through the Desire2Learn program at the University of Arizona. The proposed project provides a new sleep apnea screening framework that can addresses the need of low cost together with the need of high precision sleep apnea screening. The intellectual merit of the new sleep apnea screening framework includes a new integrated sleep monitoring wearable sensor, two deep neural network algorithms, and a smart phone based app to collect sleep data and provide screening results. The sleep monitoring wearable sensor incorporates multiple sensors on a single platform with non-volatile memory and low energy Bluetooth. The new deep learning network models classifies the severity of sleep apnea using seven levels based on the Apnea-hypopnea index (AHI). The new models include bidirectional long-short term memory layers and fully connected layers to explore and extract sensory data together with static features such as users' demographic information, existing chronic diseases, and general wellbeing. A post data processing procedure is developed to translate the probability of sleep apnea at the output of deep neural networks to sleep apnea severity using the maximum probability. A smart phone based app is also developed to execute deep neural network algorithms and display sleep apnea screening results.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.