Award for DIS at ICT Open 2019
Publication date: 01-04-2019
CWI’s DIS group has been awarded with a prize at NWO’s ICT OPEN 2019, the event for ICT research in the Netherlands. The Distributed and Interactive Systems group won the second prize for their poster “Deep Sleep. A ballistocardiographic-based Deep Learning Approach for classifying Sleep Stages." The poster was selected out of 98 posters of which 12 were nominated.
Authors: Shashank Rao (TU Delft), Abdallah El Ali (CWI), Pablo Cesar (CWI / TU Delft)
Abstract
Current techniques for tracking sleep are either obtrusive (polysomnography) or noisy with low accuracy (wearables). In this work, we aim to model a sleep classification system using an unobtrusive ballistocardiographic (BCG)-based heart sensor signal collected from a commercially available pressure-sensitive sensor sheet. We present our early work on our DeepSleep model, a hybrid deep neural network architecture comprising of CNN and LSTM layers, which is able to classify sleep stages with a mean F1-score of 74% using the BCG signal. We further employed a 2-phase training strategy to build a pre-trained model and to tackle the limited dataset size. With a classification accuracy of 83%, 77% and 63% using MIT-BIH’s ECG, Dozee’s ECG and Fitbit’s PPG datasets, respectively, we contribute early results that show promising results in such transfer learning settings. Furthermore, with a correlation coefficient of r = 0.43, our model shows a positive correlation with the SATED questionnaire perceived sleep quality scores. Although our proposed model’s performance is not yet comparable to polysomnography, we show that heart rate signals alone are a promising means for long-term sleep monitoring.
