Deep convolutional neural network-based signal quality assessment for photoplethysmogram.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Accurate bio-signal quality assessment is crucial for preventing clinical misdiagnosis.
• The rise of wearable healthcare necessitates robust methods for distinguishing physiological signals from noise.

Purpose of the Study:

• To develop and validate a signal quality assessment technology for photoplethysmogram (PPG) signals used in wearable devices.
• To ensure the reliability of PPG data collected from mobile and wearable healthcare applications.

Main Methods:

• A deep neural network (DNN) model, specifically a 1D convolutional neural network (CNN), was developed.
• The model was trained and validated using a large dataset of approximately 1.6 million PPG signal segments from the MIMIC III database.
• Model hyperparameters were optimized using Bayesian optimization.

Main Results:

• The optimized DNN model achieved high performance in classifying PPG waveform quality as 'Good' or 'Bad'.
• Key performance metrics included accuracy (0.978), specificity (0.948), sensitivity (0.993), AUC-ROC (0.985), and AUC-PR (0.980).
• The signal quality score effectively tracked decreases in pulse quality in simulated real-time applications.

Conclusions:

• The developed DNN-based model provides a reliable method for assessing PPG signal quality in wearable healthcare.
• This technology can significantly improve the accuracy of health monitoring and reduce the risk of misdiagnosis.
• The model's performance demonstrates its potential for real-time application in mobile health.