A feature exploration methodology for learning based cuffless blood pressure measurement using photoplethysmography

Area of Science:

• Biomedical Engineering
• Signal Processing
• Medical Devices

Background:

• Cuffless blood pressure measurement is crucial for continuous monitoring.
• Photoplethysmography (PPG) signals offer a non-invasive approach.
• Efficient feature extraction from PPG is vital for accurate prediction.

Purpose of the Study:

• To develop a feature exploration method for learning-based cuffless blood pressure measurement.
• To identify optimal feature sets from PPG signals for predicting systolic, diastolic, and mean blood pressure.
• To validate the performance of the proposed method against established medical standards.

Main Methods:

• Applied analytical techniques including random error elimination, adaptive outlier removal, and correlation coefficients.
• Evaluated fifty-seven feature candidates from PPG signals.
• Developed three distinct feature sets, each with eleven features, for SBP, DBP, and MBP prediction.

Main Results:

• Achieved prediction accuracies of 4.77±7.68 mmHg for SBP, 3.67±5.69 mmHg for DBP, and 3.85±5.87 mmHg for MBP.
• The proposed predictors demonstrated high accuracy on a realistic dataset.
• The feature sets were validated as light-weight and effective.

Conclusions:

• The proposed feature exploration method enables accurate cuffless blood pressure estimation from PPG signals.
• The developed predictors achieved Grade A performance according to AAMI and BHS standards.
• This approach offers a promising solution for non-invasive, continuous blood pressure monitoring.