A systematic review of non-contact medical IoT systems for physiological biomarker monitoring

Abstract

Non-contact medical Internet of Things systems measure physiological biomarkers such as heart rate (HR), respiration rate (RR), temperature, oxygen saturation (SpO₂), and blood pressure (BP) without physical contact with the patient. These systems combine sensing modalities including cameras, radar, Wi-Fi, acoustic, and infrared sensors with signal processing and learning algorithms to deliver comfortable, non-invasive monitoring. Prior reviews have largely examined individual sensing modalities or signal processing techniques in isolation, leaving cross-modality system architecture and privacy-preserving design underexamined. We conducted a systematic review following PRISMA guidelines and analyzed 88 recent studies of non-contact monitoring systems. Studies were characterized by sensing modality, system architecture, target physiological parameter, validation setting, and reported performance. We additionally assessed each study for methodological quality and examined the extent to which privacy-preserving mechanisms were incorporated into system design. Radar and vision-based systems dominate the corpus, together accounting for approximately three-quarters of included studies, while sound, Wi-Fi, embedded sensor, and mixed-modality designs remain at an earlier stage of validation. Research effort is heavily concentrated on two parameters: RR appears in 83% of studies and HR in 67%, while no other parameter exceeds 11%. BP, SpO₂, and temperature therefore remain largely unaddressed by contactless methods. Heterogeneous reporting metrics across studies prevent direct statistical comparison of performance between modalities. The findings indicate that modality selection should be driven by deployment context rather than headline accuracy. Radar suits wards and intensive care units where patients are covered or moving, vision and sound suit structured clinical spaces with stable conditions, embedded sensor designs suit eldercare beds and chairs, and mixed radar and vision systems suit precision-critical settings such as neonatal and postoperative monitoring. Persistent challenges include signal reliability in uncontrolled environments, limited benchmarking due to the absence of standardized datasets, privacy and ethical concerns, and poor integration with existing healthcare infrastructure. Building on these insights, we propose a conceptual federated, edge-based learning framework, presented as a design proposal for future implementation rather than a validated system, and outline research directions to accelerate clinical translation.