Robust Modelling of Reflectance Pulse Oximetry for SpO₂ Estimation

Area of Science:

• Biomedical Engineering
• Medical Devices
• Signal Processing

Background:

• Continuous blood oxygen saturation (SpO2) monitoring is crucial for pulmonary disorders.
• Transmittance pulse oximetry is reliable but limited to peripheral sites, problematic in cold conditions.
• Reflectance pulse oximetry offers versatile site usage and miniaturization potential but faces accuracy challenges.

Purpose of the Study:

• To develop and validate a minimally calibrated, data-driven method for accurate SpO2 estimation from reflectance photoplethysmography (PPG) waveforms.
• To address the generalizability limitations of previous machine learning approaches for reflectance pulse oximetry.

Main Methods:

• Implementation and evaluation of multiple machine learning (ML) approaches for SpO2 estimation.
• Development of a minimally calibrated, data-driven algorithm using reflectance PPG waveforms.
• Validation of the proposed method on unseen patients.

Main Results:

• Previous ML approaches failed to generalize to new patients.
• The proposed data-driven method achieved an average mean absolute error of 1.81% on unseen patients.
• The results are within the clinically permissible error margin of 2%.

Conclusions:

• The developed method provides a viable solution for accurate SpO2 estimation using reflectance pulse oximetry.
• This approach overcomes the limitations of previous ML models in terms of generalizability.
• The findings advance the understanding and application of reflectance-based pulse oximetry for patient monitoring.