Machine Learning-Based Oscillometric Blood Pressure Estimation Method

Personalized Oscillometric Blood Pressure Estimation Based on Bootstrap and Gaussian Mixture Regression

This technology uses non-parametric bootstrapping to generate pseudo-maximum amplitudes and pseudo-envelopes from a small number of blood pressure measurements, and employs Gaussian mixture models and regression to estimate subject-specific systolic and diastolic characteristic ratios to calculate blood pressure along with confidence intervals.

Blood pressure is difficult to measure accurately as it fluctuates significantly even between heartbeats due to stress, exercise, or illness. Conventional oscillometric methods apply fixed characteristic ratios uniformly, failing to account for individual differences, and often produce estimates that exceed standard regulatory error margins or lack accompanying reliability metrics.

This technology compensates for small sample sizes using bootstrapping and obtains personalized characteristic ratios through machine learning. By mapping mean and percentile pseudo-envelope points to cuff pressure, it provides both blood pressure readings and confidence intervals. Applicable to home automatic blood pressure monitors and remote health monitoring devices, it delivers highly reliable results with indicated error margins after just a few measurements.

‍

‍

Key Features:

• Obtaining an oscillometric waveform envelope from sample blood pressure measurements and acquiring pseudo-maximum amplitude and pseudo-envelopes via non-parametric bootstrapping
• Estimating individual systolic and diastolic blood pressure characteristic ratios using Gaussian mixture models and Gaussian mixture regression
• Mapping mean pseudo-envelope and 95%/5% percentile envelope points to cuff pressure to estimate blood pressure and confidence intervals
• Performing K-means clustering on feature vectors consisting of mean arterial pressure, area under the envelope, and envelope asymmetry ratio into 4 to 6 clusters

‍