Deep Learning-Based Estimation of Arterial Stiffness from PPG Spectrograms: A Novel Approach for Non-Invasive...

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Artificial Intelligence in Medicine

Background:

• Cardiovascular diseases (CVDs) are a major global health concern, with arterial stiffness being a critical risk factor.
• Non-invasive assessment of arterial stiffness, specifically Carotid-to-femoral Pulse Wave Velocity (cf-PWV), is crucial for early CVD detection and management.
• Traditional cf-PWV measurement methods can be complex and require specialized equipment.

Purpose of the Study:

• To introduce a novel, non-invasive method for estimating cf-PWV using deep learning.
• To evaluate the efficacy of photoplethysmogram (PPG) signal spectrograms as input for artificial intelligence models.
• To develop a user-friendly and accurate tool for assessing arterial stiffness.

Main Methods:

• Utilized a modified ResNet-18 deep learning architecture.
• Analyzed PPG signals from digital, radial, and brachial arteries using signal spectrograms.
• Applied the model to a simulated dataset of 4374 healthy adults.

Main Results:

• Achieved high correlation coefficients (R²): up to 0.9902 for digital, 0.9898 for radial, and 0.9825 for brachial arteries.
• Demonstrated low Mean Absolute Percentage Errors (MAPE): approximately 1.61% (digital), 1.87% (radial), and 2.08% (brachial).
• The digital artery PPG spectrograms showed particular efficacy in cf-PWV estimation.

Conclusions:

• PPG signal spectrograms are effective inputs for deep learning models to estimate cf-PWV.
• This novel approach provides an accurate, user-friendly, and non-invasive method for assessing arterial stiffness.
• The findings enhance non-invasive diagnostic capabilities for cardiovascular health.