Patch-type wearable electrocardiography and impedance pneumography for sleep staging: A multi-modal deep learning...

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Wearable Technology

Background:

• Polysomnography (PSG) is the gold standard for sleep staging but is costly and impractical for routine use.
• Accurate sleep staging is crucial for diagnosing sleep disorders and assessing sleep quality.
• There is a need for accessible, portable solutions for continuous sleep monitoring.

Purpose of the Study:

• To evaluate the feasibility of a wearable patch device using single-lead electrocardiography (ECG) and impedance pneumography (IPG) for multi-stage sleep classification.
• To compare the performance of different classification methods for sleep staging using wearable sensor data.
• To identify the most effective sensor modalities and features for accurate sleep staging.

Main Methods:

• Collected data from 92 patients using a wearable ECG-IPG device.
• Preprocessed data included bandpass filtering, segmentation, and feature extraction in time, frequency, and nonlinear domains.
• Validated three classification methods using 5-fold patient-independent cross-validation for 2-, 3-, and 4-class sleep staging tasks.

Main Results:

• The combined approach achieved 83.6% accuracy and 86.0% AUROC for 2-class (Wake/Sleep) classification.
• For 3- and 4-class tasks, feature-based methods, particularly RCNN, showed the best performance (F1-scores of 0.618 and 0.552, respectively).
• Impedance pneumography (IPG) and R-R interval (RRI) combined with motion sensors yielded the highest performance, with IPG and RRI being most effective for sleep staging.

Conclusions:

• A portable ECG-IPG system is feasible for accurate and computationally efficient sleep staging.
• The wearable device shows potential for continuous sleep monitoring and personalized health management in real-world applications.
• Feature reduction techniques (mRMR) significantly reduced training time while retaining high performance, enhancing practical usability.