Robust Prediction of Cardiorespiratory Signals from a Multimodal Physiological System on the Upper Arm

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Physiological Monitoring

Background:

• Commercial wearable sensors often use single photoplethysmography (PPG) for cardiorespiratory monitoring.
• PPG is susceptible to biases like skin tone variations and noise from motion or pressure.
• Existing single-modality systems face challenges in providing robust and accurate physiological variable estimation.

Purpose of the Study:

• To develop and evaluate a multimodal wearable device for robust cardiorespiratory monitoring.
• To assess the performance of combined physiological signals (PPG, ECG, BioZ) against single-modality limitations.
• To investigate a hierarchical approach for optimal sensor modality selection based on signal quality and subject characteristics.

Main Methods:

• A wearable device integrating multiwavelength PPG, single-sided ECG (SS-ECG), bioimpedance (BioZ), and an inertial measurement unit (IMU) was developed.
• The device was evaluated on 16 subjects for estimating heart rate (HR) and breathing rate (BR) under various noise conditions.
• A hierarchical approach was used to select optimal sensing modalities, considering skin tone and signal quality.

Main Results:

• For HR estimation, SS-ECG offered high accuracy but lower reliability, while PPG/BioZ showed lower accuracy but higher reliability, highlighting an accuracy-robustness trade-off.
• Fusing estimates from multiple modalities using ensemble bagged tree regression significantly outperformed single-modality estimates for BR.
• The multimodal approach demonstrated improved robustness against noise sources like skin tone and motion artifacts.

Conclusions:

• Multimodal cardiorespiratory monitoring using combined PPG, SS-ECG, and BioZ can overcome the accuracy-robustness trade-off inherent in single-modality systems.
• A hierarchical selection strategy enhances the reliability of wearable cardiorespiratory monitoring.
• This approach offers a more robust solution for estimating heart rate and breathing rate in diverse real-world conditions.