Using Legacy Polysomnography Data to Train a Radar System to Quantify Sleep in Older Adults and People living with Dementia

TL;DR

The paper addresses unobtrusive sleep staging in older adults and individuals with dementia using ultra-wideband radar, tackling data scarcity and PSG-to-radar domain shift. It proposes an end-to-end CNN-RNN model with adversarial domain adaptation trained on a large PSG source dataset and a small radar target dataset, enabling transfer of sleep-stage knowledge to radar signals. Key contributions include MEMD-based respiration reconstruction, label-aware domain alignment, and robust evaluation across radar placements and health conditions, achieving about 79.5% accuracy and a 0.65 Cohen's kappa on radar data. These results demonstrate the viability of UWB radar for nonintrusive, home-based sleep assessment in dementia care, with implications for continuous monitoring and care planning.

Abstract

Objective: Ultra-wideband radar technology offers a promising solution for unobtrusive and cost-effective in-home sleep monitoring. However, the limited availability of radar sleep data poses challenges in building robust models that generalize across diverse cohorts and environments. This study proposes a novel deep transfer learning framework to enhance sleep stage classification using radar data. Methods: An end-to-end neural network was developed to classify sleep stages based on nocturnal respiratory and motion signals. The network was trained using a combination of large-scale polysomnography (PSG) datasets and radar data. A domain adaptation approach employing adversarial learning was utilized to bridge the knowledge gap between PSG and radar signals. Validation was performed on a radar dataset of 47 older adults (mean age: 71.2), including 18 participants with prodromal or mild Alzheimer disease. Results: The proposed network structure achieves an accuracy of 79.5% with a Kappa value of 0.65 when classifying wakefulness, rapid eye movement, light sleep and deep sleep. Experimental results confirm that our deep transfer learning approach significantly enhances automatic sleep staging performance in the target domain. Conclusion: This method effectively addresses challenges associated with data variability and limited sample size, substantially improving the reliability of automatic sleep staging models, especially in contexts where radar data is limited. Significance: The findings underscore the viability of UWB radar as a nonintrusive, forward-looking sleep assessment tool that could significantly benefit care for older people and people with neurodegenerative disorders.

Figures (13)

• Figure 1: Overview of the proposed approach for radar-based sleep staging and transfer learning. The nighttime activity signal and respiration signal are used to predict the sleep stage and a domain adaptation model is used to align feature distributions between the source domain PSG data and the target domain radar signals.
• Figure 2: The setup of the experiment: Radar recording system (Fig A&B) and floor map of the UK DRI clinical research facility at the University of Surrey (Fig C&D). The position of the radar is marked with a red pentagram. Figure reuse from yin2025unobtrusive.
• Figure 3: Signal pre-processing and respiration reconstruction pipeline. A: initial range profile of the raw radar signal, and B: range profile after clutter removal. C: Radar-derived and IMU-derived activity signal. D: Phase-demodulated range–time map within the participant’s range of interest. E: 30-second comparison between radar $R(t)$ and belt data.
• Figure 4: Overnight respiration rate comparison between radar and PSG belt channel, and comparison of overnight activity signals as detected by radar and IMU.
• Figure 5: Model overview. The end-to-end network comprises a feature extractor $\mathcal{F}$, a sleep-stage classifier $\mathcal{C}$, and a domain discriminator $\mathcal{D}$.