Using machine learning models for cuffless blood pressure estimation with ballistocardiogram and impedance...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Blood pressure (BP) monitoring is critical for managing hypertension and cardiovascular diseases.
• Conventional cuff-based sphygmomanometers are incompatible with mobile health (mHealth) applications.
• Cuffless BP measurement offers a promising alternative for continuous and convenient monitoring.

Purpose of the Study:

• To develop and validate a cuffless BP estimation method using ballistocardiograms (BCG) and impedance plethysmograms (IPG) from a weight-fat scale.
• To utilize deep learning and machine learning for accurate signal quality classification and BP parameter extraction.
• To assess the feasibility of integrating this technology into mHealth devices.

Main Methods:

• A stacked model combining 1D Convolutional Neural Network (1D CNN) and Gated Recurrent Unit (GRU) was used for BCG and IPG signal quality classification.
• Pulse transit time (PTT) parameters were extracted from BCG and IPG signals.
• Random Forest (RF) and XGBoost models were employed to estimate systolic BP (SBP) and diastolic BP (DBP) using PTT parameters and heart rate (HR).

Main Results:

• Signal quality classification achieved an accuracy of 0.989.
• Five-fold cross-validation yielded high Pearson correlation coefficients: 0.953 ± 0.007 for SBP and 0.935 ± 0.007 for DBP.
• The XGBoost model demonstrated low mean absolute differences: 3.54 ± 0.34 mmHg for SBP and 2.57 ± 0.17 mmHg for DBP.

Conclusions:

• The proposed method significantly enhances the accuracy of cuffless BP measurement.
• Integration into weight-fat scales offers an unconstrained device for mHealth applications.
• This technology holds potential for improved remote patient monitoring and management of cardiovascular health.