Acoustical detection of coronary occlusions using neural networks

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Coronary artery disease (CAD) diagnosis relies on invasive and noninvasive methods.
• Noninvasive acoustic analysis of heart sounds offers a promising alternative for CAD detection.
• Developing accurate noninvasive diagnostic tools for CAD is crucial for early intervention.

Purpose of the Study:

• To develop and evaluate a nonlinear neural network classifier for noninvasive acoustic detection of coronary artery disease.
• To assess the diagnostic performance of the neural network using features derived from diastolic heart sounds.
• To compare the efficacy of this novel approach with existing noninvasive diagnostic methods for CAD.

Main Methods:

• A nonlinear neural network classifier was designed, incorporating a feature vector derived from diastolic heart sounds.
• Linear prediction coefficients from an autoregressive method, following adaptive line enhancement, formed the input pattern for the neural network.
• The network was trained using the backpropagation algorithm on a dataset of 112 patient recordings (70 abnormal, 42 normal).

Main Results:

• The neural network correctly identified 50 out of 64 patients with coronary artery disease.
• It also correctly identified 32 out of 36 patients without coronary artery occlusions.
• The classifier demonstrated a high capability in distinguishing between normal and abnormal patients.

Conclusions:

• The developed nonlinear neural network classifier is effective for noninvasive acoustic detection of coronary artery disease.
• This approach exhibits superior diagnostic capability compared to other available noninvasive methods for CAD.
• Noninvasive acoustic analysis combined with advanced machine learning offers a powerful tool for CAD diagnosis.