Cuff-Less Blood Pressure Prediction Based on Photoplethysmography and Modified ResNet.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Cardiovascular Health

Background:

• Cardiovascular disease (CVD) prevalence and mortality are increasing globally.
• Blood pressure (BP) is a critical indicator for CVD prevention and treatment.
• Current cuff-based BP monitoring is intermittent and inconvenient.

Purpose of the Study:

• To develop a continuous BP monitoring system using photoplethysmography (PPG) signals.
• To leverage a ResNet34 deep learning framework for accurate BP prediction.
• To validate the feasibility of cuff-free BP monitoring for wearable technology.

Main Methods:

• Utilized a ResNet34 deep learning network for BP prediction from PPG signals.
• Implemented multi-scale feature extraction and channel attention modules for enhanced accuracy.
• Employed the Huber loss function for stable model training and optimization.
• Validated the model on a subset of the MIMIC dataset.

Main Results:

• Predicted BP errors met AAMI standards for both systolic (SBP) and diastolic (DBP).
• Achieved Grade A accuracy for DBP and near Grade A for SBP according to BHS standards.
• Demonstrated the potential of PPG signals and deep learning for continuous BP monitoring.

Conclusions:

• The proposed deep learning method shows significant potential for continuous, cuff-free BP monitoring.
• The technology is suitable for integration into portable and wearable devices like smartwatches.
• This approach aligns with future trends in personalized and remote health monitoring.