Integrating transfer learning with scalogram analysis for blood pressure estimation from PPG signals

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiovascular Health Monitoring

Background:

• Continuous blood pressure (BP) monitoring is vital for cardiovascular health assessment and early detection of heart disorders.
• Elevated blood pressure is a key indicator of potential heart complications, necessitating reliable monitoring methods.

Purpose of the Study:

• To develop and validate a deep learning-based approach for accurate blood pressure estimation using photoplethysmography (PPG) signals.
• To explore the efficacy of various deep learning models integrated with transfer learning for non-invasive BP measurement.

Main Methods:

• Photoplethysmogram (PPG) signals were transformed into scalograms using Continuous Wavelet Transform (CWT).
• Six deep learning models (VGG16, ResNet50, InceptionV3, NASNetLarge, InceptionResNetV2, ConvNeXtTiny) were employed for deep feature extraction.
• Extracted features were used with a Random Forest regression model for BP estimation.
• Performance was evaluated using Mean Absolute Error (MAE) and Standard Deviation (SD), adhering to AAMI and BHS clinical standards.

Main Results:

• ConvNeXtTiny and VGG16 models exhibited strong performance in BP estimation.
• ConvNeXtTiny achieved an MAE of 2.95 mmHg and SD of 4.11 mmHg for systolic BP, and MAE of 1.66 mmHg and SD of 2.60 mmHg for diastolic BP.
• The results align with established clinical standards (AAMI, BHS), indicating high reliability.

Conclusions:

• Deep learning and transfer learning enable reliable blood pressure estimation from PPG signals.
• The ConvNeXtTiny model provides a dependable, non-invasive method for continuous BP monitoring, meeting clinical requirements.
• This approach can significantly enhance cardiovascular health monitoring and facilitate early detection of cardiac issues.