Imbalanced feature generation based on bootstrap power spectral curve for estimating respiratory rate

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Cardiorespiratory Monitoring

Background:

• Rapid respiratory rate (RR) changes in older adults signal potential serious illness, necessitating accurate estimation for cardiorespiratory fitness assessment.
• Machine learning algorithms struggle with data imbalance, where certain respiratory patterns (normal breathing) have larger sample sizes than others (dyspnea, hypopnea), leading to significant errors.
• Existing methods are often unsuitable for medical decision-making due to these data imbalance-induced inaccuracies.

Purpose of the Study:

• To introduce a novel methodology combining bootstrap-based imbalanced feature generation (BIFG) with Gaussian processes for accurate RR estimation and uncertainty quantification.
• To address the critical issue of data imbalance in respiratory rate monitoring, particularly concerning underrepresented patterns like dyspnea and hypopnea.
• To enhance the reliability of machine learning models for medical decision-making in cardiorespiratory health.

Main Methods:

• Developed a novel methodology integrating bootstrap-based imbalanced feature generation (BIFG) with Gaussian processes.
• Utilized parametric bootstrap models to generate artificial feature curves, specifically targeting underrepresented data segments (dyspnea and hypopnea).
• Employed a non-parametric bootstrap approach to increase artificial feature curve generation, selectively applying them to imbalanced data portions for improved model training.

Main Results:

• The proposed BIFG methodology effectively addresses data imbalance in RR estimation.
• Achieved accurate prediction of complex nonlinear relationships between photoplethysmography signal features and reference RR.
• Demonstrated superior predictive performance and uncertainty quantification, with mean absolute errors of 0.89 and 1.44 beats per minute across two datasets.

Conclusions:

• The BIFG approach significantly improves the accuracy and reliability of RR estimation in older adults, even with imbalanced datasets.
• This methodology offers a robust solution for machine learning in cardiorespiratory monitoring, enhancing its utility in medical decision-making.
• The accurate estimation of RR and its associated uncertainty provides valuable insights for assessing cardiorespiratory fitness and detecting serious illness.