Oscillometric blood pressure estimation using machine learning-based mapping of waveform features.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Cardiovascular Health

Background:

• Hypertension detection and monitoring are crucial for managing cardiovascular diseases like heart attacks and strokes.
• Traditional blood pressure measurement methods have limitations in accuracy and convenience.
• Accurate and sensitive non-invasive blood pressure monitoring is needed.

Purpose of the Study:

• To propose a novel hybrid classification-mapping model for estimating Systolic (SBP) and Diastolic (DBP) blood pressure.
• To explore new features from oscillometric waveforms (OW) for improved BP estimation.
• To optimize machine learning classifiers using feature ranking techniques.

Main Methods:

• Utilized a hybrid classification-mapping model on 155 subjects from the University of New South Wales NIBP dataset.
• Extracted novel beat-related features from oscillometric waveforms.
• Employed eight feature ranking techniques to optimize KNN, Ensemble KNN, Ensemble Bagged Tree, and SVM classifiers.

Main Results:

• Achieved a Mean Absolute Error (MAE) of 1.28 ± 2.27 mmHg for SBP estimation.
• Demonstrated comparable results to existing methods for DBP estimation.
• Optimized machine learning models through feature selection and ranking.

Conclusions:

• The proposed hybrid model offers a more accurate and sensitive non-invasive method for blood pressure estimation.
• This approach could facilitate reliable and convenient remote blood pressure monitoring.
• Implementation may enhance early detection and management of hypertension and its consequences.