Predicting Respiratory Rate from Electrocardiogram and Photoplethysmogram Using a Transformer-Based Model.

Area of Science:

• Biomedical Engineering
• Physiological Monitoring
• Machine Learning in Healthcare

Background:

• Respiratory rate (RR) is crucial for diagnosis and prognosis but difficult to measure accurately.
• Manual RR counting is standard but prone to errors.
• Existing algorithms for predicting RR from ECG and PPG signals have limited accuracy and generalization.

Purpose of the Study:

• To develop an enhanced Transformer model (TransRR) for accurate RR prediction using ECG and PPG signals.
• To evaluate the model's generalization capability on unseen data.
• To introduce a novel preprocessing pipeline to boost model performance.

Main Methods:

• Developed an enhanced Transformer model incorporating inception blocks.
• Utilized subject-level ten-fold cross-validation on BIDMC and CapnoBase datasets.
• Implemented a new preprocessing pipeline for ECG and PPG signals.

Main Results:

• Achieved superior performance compared to five popular deep-learning methods.
• Reduced mean absolute error by 36.5% (MAE=1.2).
• Increased correlation coefficient by 84.8% (R=0.85) on the test set.

Conclusions:

• The TransRR model demonstrates significant improvements in RR prediction accuracy and generalization.
• The proposed preprocessing pipeline further enhances model performance.
• TransRR is expected to accelerate the clinical adoption of automated RR estimation.