Non-invasive tidal volume estimation with wearable sensors using a high-gain observer and deep learning

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Respiratory Physiology

Background:

• Non-invasive respiratory monitoring is crucial for patient care.
• Traditional spirometry is accurate but invasive and limits mobility.
• Wearable sensors offer a promising alternative for continuous monitoring.

Purpose of the Study:

• To develop and validate a non-invasive method for tidal volume (TV) estimation using wearable inertial measurement unit (IMU) sensors.
• To integrate a nonlinear high-gain observer (HGO) with a convolutional neural network long short-term memory (CNN-LSTM) network for enhanced TV prediction.
• To assess the accuracy and robustness of the proposed method against sensor placement variations.

Main Methods:

• Utilized wearable IMU sensors to collect thoracoabdominal displacement data.
• Employed a nonlinear high-gain observer (HGO) to process IMU data, mitigating sensor drift and gravity effects.
• Integrated HGO-processed data and raw IMU signals as inputs for a CNN-LSTM deep learning network.
• Trained and validated the CNN-LSTM model on data from 6 subjects in an IRB-approved study.

Main Results:

• The integrated HGO-CNN-LSTM model achieved a high degree of accuracy in estimating tidal volume.
• The system demonstrated robustness to variations in sensor placement, even after removal and re-wearing.
• An average Root Mean Square (RMS) error of 40.38 mL was recorded in experimental trials.
• The non-invasive method showed potential for replacing conventional spirometry.

Conclusions:

• The developed wearable sensor system coupled with HGO and CNN-LSTM algorithms provides accurate and robust non-invasive tidal volume estimation.
• This technology holds significant potential for continuous respiratory monitoring in clinical and remote settings.
• The findings suggest a viable alternative to invasive spirometry, enhancing patient comfort and mobility.