A learning-based image processing approach for pulse wave velocity estimation using spectrogram from peripheral pulse...

Area of Science:

• Cardiovascular Physiology
• Biomedical Signal Processing
• Machine Learning in Healthcare

Background:

• Carotid-to-femoral pulse wave velocity (cf-PWV) is a key indicator of arterial stiffness and cardiovascular disease.
• Traditional cf-PWV measurement is invasive and prone to errors.
• A non-invasive peripheral method is needed to simplify assessment and improve patient care.

Purpose of the Study:

• To propose a novel, non-invasive methodology for estimating cf-PWV using peripheral signals.
• To evaluate the effectiveness of spectrogram representation combined with machine learning for cf-PWV estimation.
• To compare different feature extraction techniques for this novel approach.

Main Methods:

• Utilized spectrogram representation of peripheral pulse wave signals (PPG or BP).
• Employed three feature extraction methods: semi-classical signal analysis (SCSA), Law's mask, and central statistical moments.
• Integrated extracted features with machine learning models (MLP) for cf-PWV prediction.

Main Results:

• Achieved R² ≥ 0.90 for noise-free signals using the MLP model.
• Demonstrated robust performance with R² ≥ 0.91 even with added noise, using SCSA features and MLP.
• Spectrogram representation proved effective across different feature extraction methods and signal types.

Conclusions:

• The proposed methodology enables accurate and non-invasive cf-PWV estimation from peripheral signals.
• Spectrogram analysis combined with machine learning offers a promising alternative to traditional methods.
• Further validation with in-vivo signals is warranted.