CapNet: A Deep Learning-based Framework for Estimation of Capnograph Signal from PPG.

Area of Science:

• Biomedical Engineering
• Signal Processing

Background:

• Continuous respiratory monitoring is crucial for patients with respiratory deficiencies.
• Traditional capnography is invasive and costly, necessitating alternative methods.
• Photoplethysmogram (PPG) signals offer a promising, less expensive approach for respiratory monitoring.

Purpose of the Study:

• To introduce CapNet, a novel deep learning framework for estimating capnograph signals from PPG signals.
• To evaluate CapNet's performance against traditional methods and existing deep learning models.

Main Methods:

• Developed a deep learning framework (CapNet) utilizing PPG signals as input.
• Trained, validated, and tested CapNet using the IEEE TMBE Respiratory Rate Benchmark dataset.
• Compared CapNet's performance with two traditional signal processing algorithms and the RespNet deep neural network.

Main Results:

• CapNet demonstrated superior performance with lower Mean Squared Error (MSE) and higher cross-correlation values.
• The proposed framework achieved better results than traditional methods and the RespNet model.
• The study validates the effectiveness of deep learning for respiratory signal extraction from PPG.

Conclusions:

• CapNet provides a feasible and implementable solution for continuous respiratory monitoring.
• This non-invasive approach using PPG signals can significantly benefit patients with respiratory ailments.
• Deep learning-based PPG analysis presents a viable alternative to traditional capnography.