Sensors and computational methods for sleep quality assessment: A systematic review

Background Wearable, bed-based, and non-contact sensor systems combined with computational models are increasingly used as less intrusive and more scalable alternatives to polysomnography for sleep monitoring. Objective This systematic review evaluates sensor technologies, extracted physiological and behavioral features, computational methods, validation approaches, and practical limitations in sleep-quality assessment. Unlike reviews focused on a single sensor or clinical task, this review jointly maps sensor modalities to computational methods, target sleep outcomes, reference standards, and readiness for home or clinical deployment. Methods A systematic search was conducted in IEEE Xplore, PubMed, Springer, Elsevier/ScienceDirect, and MDPI. Studies published between January 2014 and May 2025 were included when they assessed sleep quality or related sleep outcomes using wearable, flexible, bed-based, non-contact, or non-invasive sensors together with computational or signal-processing methods. A total of 57 publications were considered, comprising 54 primary studies and three contextual review articles. Results Among the 54 primary studies, wearable sensor studies represented the largest category, accounting for 16 studies, followed by non-contact sleep monitoring and sleep-apnea detection with 11 studies each, advanced computational models with 10 studies, and sleep-posture recognition with 6 studies. Accelerometry, actigraphy, pressure sensing, photoplethysmography, electrocardiography, ballistocardiography, radar, and acoustic sensing were the principal modalities. Frequently extracted features included motion activity, heart-rate variability, respiratory patterns, pressure distribution, acoustic signals, and posture dynamics. Classical machine-learning methods remained useful for smaller datasets and interpretable features, whereas deep-learning models were increasingly applied to multimodal, spatial, and temporal signals. However, performance was strongly influenced by the target task, population, dataset, reference standard, and validation protocol. Conclusion Wearable and non-contact technologies show substantial potential for longitudinal sleep monitoring, apnea screening, sleep-wake estimation, and posture recognition. Nevertheless, small cohorts, heterogeneous protocols, inconsistent polysomnography-based validation, and limited external testing currently prevent their general acceptance as equivalent replacements for clinical polysomnography.