Estimating changes in systolic blood pressure based on pulse wave morphology using paired segment comparison

Area of Science:

• Biomedical Engineering
• Physiological Monitoring
• Machine Learning in Healthcare

Background:

• Continuous non-invasive blood pressure (NIBP) monitoring is limited, as cuff-based methods provide intermittent readings and miss short-term systolic blood pressure (SBP) fluctuations.
• Existing photoplethysmogram (PPG)-based SBP estimation methods often require calibration due to subject-specific vascular properties, limiting their generalizability.
• There is a need for accurate, continuous, and non-invasive methods to monitor blood pressure, especially during sleep or when cuff inflations are disruptive.

Purpose of the Study:

• To develop and evaluate a learning-based approach for estimating the change in systolic blood pressure (ΔSBP) using PPG signals.
• To overcome the limitations of current PPG-based SBP estimation by focusing on relative changes rather than absolute values.
• To assess the performance and stability of the proposed ΔSBP estimation method across diverse patient datasets and over extended periods.

Main Methods:

• A learning-based algorithm was implemented to compare pairs of 20-second PPG segments to estimate ΔSBP.
• The approach was developed and validated on a large cohort of 12,279 patients from the VitalDB and MIMIC-III datasets, with a held-out test set of 2,457 patients.
• Performance was quantified using Root Mean Squared Error (RMSE), Pearson correlation coefficient (r), and Cohen's kappa (κ).

Main Results:

• The PPG-based ΔSBP estimation achieved an RMSE of 14.53/11.37 mmHg and a correlation (r) of 0.79/0.65 on the VitalDB/MIMIC-III test sets.
• Model predictions showed a stronger correlation with SBP than heart rate, indicating reliance on pulse wave morphology.
• Stable ΔSBP estimation was demonstrated over a 28-hour period, with errors primarily occurring when waveform changes were minimal despite significant ΔSBP.

Conclusions:

• The study supports the potential of morphology-based relative blood pressure estimation using PPG signals.
• The developed approach can potentially complement cuff-based monitoring by providing continuous SBP trend data between intermittent measurements.
• This method could be particularly valuable during sleep or in situations where cuff inflations are undesirable or impractical.