Development of a Personalized Multiclass Classification Model to Detect Blood Pressure Variations Associated with...

Area of Science:

• Biomedical Engineering
• Physiological Monitoring
• Machine Learning Applications

Background:

• Continuous blood pressure monitoring is crucial for understanding regulatory mechanisms.
• Personalized machine learning offers a novel approach for tracking blood pressure fluctuations.
• Photoplethysmography (PPG) signals contain valuable data for physiological assessments.

Purpose of the Study:

• To develop a personalized machine learning model for blood pressure tracking.
• To evaluate the model's performance based on training set size and personalization.
• To assess the model's ability to detect pulse waveform variations under different workloads.

Main Methods:

• Extracted data-driven photoplethysmograph features from brachial and digital arteries.
• Utilized a random forest classifier trained on data from 28 healthy subjects.
• Investigated model behavior with varying training set sizes and personalization levels.

Main Results:

• Achieved aggregated accuracy of 95.1%, precision of 95.2%, recall of 95%, and F1-score of 95.4%.
• Optimal performance was observed when 30% of a subject's pulse waveforms were combined with data from five source subjects.
• The pre-training stage enabled differentiation of beat-to-beat pulse waveform morphology under cognitive and physical stress.

Conclusions:

• A personalized machine learning model can effectively track blood pressure using PPG features.
• Combining subject-specific data with external data enhances model accuracy and robustness.
• The developed system shows promise for non-invasive, continuous blood pressure monitoring and workload assessment.