Estimation of Vascular Stenosis and Blood Flow Rate Using Machine Learning Classification of Blood Sounds

Area of Science:

• Biomedical Engineering
• Medical Acoustics
• Signal Processing

Background:

• Phonoangiography (PAG) quantifies blood sounds (bruits) noninvasively.
• Accurate detection of vascular stenosis is crucial for preventing complications.

Purpose of the Study:

• To develop and validate a vascular phantom for in vitro bruit reproduction.
• To assess the efficacy of machine learning models for stenosis classification using PAG features.
• To evaluate PAG's potential for predicting flow rate and vascular screening.

Main Methods:

• Developed a vascular phantom with blood-mimicking fluid and scattering media.
• Recorded acoustic signals using clinical stethoscopes at multiple sites.
• Extracted descriptive features and applied machine learning for stenosis classification and flow rate prediction.

Main Results:

• Machine learning models achieved up to 100% accuracy in stenosis classification.
• Delta feature analysis improved detection accuracy and reduced site selection sensitivity.
• PAG features successfully predicted flow rate, demonstrating screening feasibility.

Conclusions:

• PAG is a feasible, cost-effective, real-time method for early stenosis detection and hemodynamic evaluation.
• This approach may enable point-of-care or at-home detection of hemodynamically significant vascular stenosis.