Synthetic photoplethysmogram (PPG) signal generation using a genetic programming-based generative model.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Photoplethysmograph (PPG) technology is increasingly integrated into smart devices for cardiac monitoring.
• Developing generative models for synthetic PPG signals faces challenges with data diversity and limited training datasets.
• Deep learning models require extensive and varied data for effective PPG signal generation.

Purpose of the Study:

• To propose a novel generative model for synthetic PPG signal generation.
• To address the limitations of data diversity and availability in current PPG signal generation methods.
• To leverage genetic programming (GP) for creating accurate and varied PPG data.

Main Methods:

• A genetic programming (GP) approach was adopted to create a generative model.
• The GP model automatically determines mathematical model structures and combinations.
• The model utilizes an initial PPG signal sample to generate diversified and accurate data.

Main Results:

• The proposed GP model achieved a mean square error (MSE) of 0.0001.
• Root mean square error (RMSE) was recorded at 0.01, indicating high accuracy.
• A correlation coefficient of 0.999 demonstrated strong agreement with real PPG signals.
• The GP approach outperformed conventional methods in generating synthetic PPG data.

Conclusions:

• The GP-based generative model is effective for producing diverse and accurate synthetic PPG signals.
• The approach overcomes challenges associated with limited data and enhances data diversity.
• The model demonstrates efficiency and applicability, particularly in resource-constrained environments for cardiac monitoring.