Smartphone-based and non-invasive sleep stage identification system with piezo-capacitive sensors Pérez Ávila, Antonio Javier Ruiz Herrera, Noelia Martínez Olmos, Antonio Carvajal Rodríguez, Miguel Ángel Capitán Vallvey, Luis Fermín López Ruiz, Nuria Palma López, Alberto José Piezo-capacitive sensors PVDF Sleep monitoring Smartphone Machine learning Authors would like to express their gratitude to LoMonaco and Dr. Alejandro Guillén Riquelme for their advisory support. This research was funded by Junta de Andalucía (Spain) under project PYC20-RE-040 UGR and by Spanish MCIN/AEI/10.13039/501100011033/ with project PID2022–138727OB-I00. The project was partially supported by European Regional Development Funds “ERDF A way of making Europe”. Moreover, we would like to thank the people managing the Sleep Laboratory of the University of Granada for allowing the use of equipment when needed. A non-invasive, wireless, smartphone-based electronic measurement system for sleep stage identification is presented in this work. Ballistocardiograph signals are collected by two piezo-capacitive thin film strips located on the mattress base. Suitable analog conditioning circuits and digital pre-processing techniques are applied to obtain the heart and breathing rates (HR, BR), and an activity index (ACT) related to the body movements during the sleep. An initial calibration stage is proposed where analog signal amplification is fitted to each subject, from which activity index is derived. Features considered for machine learning classifications were the mentioned data and the time variabilities of HR and BR represented by the features R(k) and B(k), respectively. Support Vector Machine (SVM) and K-Nearest-Neighbour (KNN) classifiers are employed in both flat and hierarchical classifi- cation scenarios for Wake – Non Rapid Eye Movement – Rapid Eye Movement (WAKE/NREM/REM) sleep stage identification. Twelve healthy subjects were recorded with the developed system using a polysomnograph (PSG) as reference data. When compared with PSG, the presented system achieved an average accuracy of 69 % using only three features: R(k), B(k), and ACT, highlighting an 88.2 % recall for NREM stage identification. These findings suggest that accounting only for time variability features and activity, satisfactory results can be pro- vided as a complementary alternative for sleep stage identification, with a smartphone-based electronic system designed as an affordable, versatile, and simple tool for household applications. 2024-07-04T11:43:35Z 2024-07-04T11:43:35Z 2024-10-01 journal article A.J. Pérez-Ávila et al. Smartphone-based and non-invasive sleep stage identification system with piezo-capacitive sensors. Sensors & Actuators: A. Physical 376 (2024) 115659. https://doi.org/10.1016/j.sna.2024.115659 https://hdl.handle.net/10481/92981 10.1016/j.sna.2024.115659 eng open access Elsevier