Non-Invasive Detection of Coronary Artery Disease and Valvular Disorders Using a Multichannel PCG Vest

Area of Science:

• Cardiology and Biomedical Engineering
• Signal Processing and Machine Learning

Background:

• Cardiovascular diseases (CVD), including coronary artery disease (CAD) and valvular heart disorder (VHD), are leading global causes of mortality.
• Current screening methods for CVD can be invasive or lack the capacity for mass-scale detection.
• Phonocardiogram (PCG) signals offer a non-invasive acoustic signature that is altered by CAD and VHD.

Purpose of the Study:

• To develop and evaluate a multiclass classification model for differentiating between CAD, VHD, and normal heartbeats.
• To assess the efficacy of using a wearable multichannel PCG vest for CVD detection.
• To establish a novel, non-invasive screening tool for various cardiovascular conditions.

Main Methods:

• Implementation of a multiclass classification model using Linear Frequency Cepstral Coefficients (LFCCs).
• Utilized a Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel for classification.
• Collected PCG signals from subjects using an innovative wearable multichannel PCG vest.

Main Results:

• The multiclass classification model achieved a subject-level accuracy of 81.53% for differentiating CAD, VHD, and normal heartbeats.
• High sensitivity was reported: 85.20% for VHD and 81.47% for CAD.
• A binary classification task distinguishing normal from abnormal heartbeats (CAD/VHD) yielded an accuracy of 82.01%.

Conclusions:

• This study demonstrates the feasibility of using PCG signals and machine learning for non-invasive CVD detection.
• The developed model shows promise as an effective screening tool for distinguishing between CAD, VHD, and normal heartbeats.
• This research is the first to apply multiclass classification across different CVD categories using PCG data from a single hardware source.