Optimizing the input feature sets and machine learning algorithms for reliable and accurate estimation of continuous,...

Area of Science:

• Biomedical Engineering
• Medical Devices
• Machine Learning in Healthcare

Background:

• Continuous blood pressure (BP) monitoring is increasingly important with the rise of mobile health and wearables.
• Current cuffless BP devices lack the accuracy and reliability required for clinical use.
• Non-interventional BP measurement methods are highly sought after for patient convenience and continuous monitoring.

Purpose of the Study:

• To develop and validate a method for accurate, non-interventional continuous blood pressure estimation using cuffless devices.
• To investigate the efficacy of combining multimodal physiological and demographic data with machine learning algorithms for BP prediction.
• To assess the performance of the developed method against established clinical standards.

Main Methods:

• Utilized multimodal feature datasets including pulse arrival time (PAT), pulse wave morphology (PWM), and demographic data.
• Employed optimized Machine Learning (ML) algorithms for the estimation of Systolic BP (SBP), Diastolic BP (DBP), and Mean Arterial Pressure (MAP).
• Validated the accuracy against the gold standard Intra-Arterial BP, adhering to IEC/ANSI 80601-2-30 (2018) standards.

Main Results:

• The multimodal approach achieved an accuracy within a 5 mmHg bias of Intra-Arterial BP for SBP, DBP, and MAP, meeting clinical standards.
• Diastolic BP calculations from 126 datasets of hemodynamically compromised patients showed a standard deviation within 8 mmHg.
• Statistical analysis revealed significant differences among ML algorithms but no significant differences among the multimodal feature datasets.

Conclusions:

• Optimized ML algorithms combined with multimodal features can enable reliable and accurate continuous BP estimation in cuffless devices.
• The findings suggest a viable pathway for developing cuffless BP monitors that meet clinical adoption requirements.
• Further development using real-world data (RWD) could accelerate the clinical integration of advanced cuffless BP monitoring technology.