2021 Fiscal Year Research-status Report
New ubiquitous computing system uniting chaos theory and data science for sleep apnea hypopnea syndrome (SAHS) screening with wearable devices
| Project/Area Number |
21K17670
|
|---|
| Research Institution | Kyoto University of Advanced Science |
|---|
Principal Investigator |
梁 滋路 京都先端科学大学, 工学部, 講師 (10782807)
|
|---|
| Project Period (FY) |
2021-04-01 – 2025-03-31
|
| Keywords | wearable computing / sleep apnea / ubiquitous computing / personal informatics / digital health / mHealth |
|---|
| Outline of Annual Research Achievements |
This study aims to develop novel algorithms for SAHS screening using consumer wearable devices. Our prior studies revealed that consumer wearables were inaccurate in measuring sleep structure, and existing clinical sleep staging models are not applicable to consumer wearables as they rely on a different set of input signals. Therefore, the work in the first FY was focused on crafting accurate sleep staging models for consumer wearables. We developed original sleep staging models that work for two most popular wearable wristbands, i.e., Fitbit and Apple Watches, respectively. We also thoroughly investigated the properties of the models under various conditions. The best average performance on Cohen's Kappa achieved by the models was 0.43 for Fitbit and 0.35 for Apple Watches.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
While initially the model development was planned to be surrounding Fitbit, eventually we were able to create models for Apple Watches as well. By accommodating the two most widely used consumer wearables, the SAHS screening system (to be developed later) will be able to reach a wider audience. We were also the first to thoroughly investigate the properties of the models with varied parameters, including epoch size, feature set, resampling methods, and machine learning techniques. We found that when using large feature set, model performance was heavily affected by resampling and epoch sizes. This phenomenon was not observed when small feature set was used. Such investigation allowed us to gain additional insights into the strength and limitations of the models under various conditions.
|
| Strategy for Future Research Activity |
The outcome of the first phase lays a solid foundation for the subsequent phases of developing the proposed SAHS screening algorithms and system. As planned in the original proposal, the next step includes conducting a pilot data collection experiment with concurrent use of medical and consumer wearables. Data science techniques and chaos theory will be used together to identify features that have strong statistical power in detecting SAHS versus healthy controls.
|
| Causes of Carryover |
The incurring amount is due to the availability of a new open-access dataset (https://www.kaggle.com/datasets/msarmi9/walch-apple-watch-sleep-dataset) that saved the trouble of conducting one of the planned data collection experiments in the first FY. The incurring amount will be used to cover the cost occurred during the experiments in the subsequent years and the participants incentives.
|
