PPG Signals-Based Blood-Pressure Estimation Using Grid Search in Hyperparameter Optimization of CNN-LSTM

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Continuous noninvasive blood-pressure measurement (cNIBP) using photoplethysmography (PPG) is crucial for patient monitoring.
• Existing cNIBP systems require performance improvements in accuracy and precision for reliable health management.

Purpose of the Study:

• To propose a robust blood-pressure estimation system using a convolutional long short-term memory (CNN-LSTM) neural network.
• To enhance the extraction of meaningful information from PPG signals for improved blood pressure monitoring.
• To reduce hyperparameter optimization complexity through a grid-search approach.

Main Methods:

• Utilized the MIMIC III dataset containing PPG and arterial-blood-pressure (ABP) signals.
• Implemented a CNN-LSTM model with five convolutional layers, one LSTM layer, and two fully connected layers.
• Employed five-fold cross-validation and grid search for optimal hyperparameter selection and model validation.

Main Results:

• Achieved a standard deviation (SD) of 7.89 ± 3.79 mmHg for systolic blood pressure (SBP) estimation.
• Attained a mean absolute error (MAE) of 5.34 ± 2.89 mmHg for diastolic blood pressure (DBP) estimation.
• Demonstrated satisfactory performance meeting BHS, AAMI, and IEEE standards for blood pressure monitoring devices.

Conclusions:

• The proposed CNN-LSTM model offers a robust and accurate solution for continuous noninvasive blood pressure estimation.
• The grid-search method effectively optimized the CNN-LSTM hyperparameters, improving PPG signal analysis.
• The system meets established international standards, indicating its potential for clinical application in blood pressure monitoring.