Intelligent Phonocardiography for Screening Ventricular Septal Defect Using Time Growing Neural Network

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Distinguishing Ventricular Septal Defect (VSD) from atrioventricular valve regurgitation can be challenging using traditional methods.
• Phonocardiography (PCG) offers a non-invasive method for cardiac auscultation, but interpretation requires expertise.
• Intelligent systems may enhance the diagnostic capabilities of PCG analysis.

Purpose of the Study:

• To evaluate the effectiveness of an intelligent phonocardiography system in differentiating between VSD and atrioventricular valve regurgitation.
• To introduce and assess a novel machine learning algorithm for classifying cardiac sound recordings.
• To determine the diagnostic performance of this AI-driven approach in a pediatric cohort.

Main Methods:

• Development and application of a Time Growing Neural Network (TGNN) machine learning model.
• Classification of phonocardiographic recordings from 90 pediatric participants (30 VSD, 30 valvular regurgitation, 30 healthy controls).
• Informed consent obtained from all participants prior to data collection.

Main Results:

• The intelligent phonocardiography system achieved an overall accuracy of 86.7%.
• The sensitivity of the method for detecting the targeted conditions was calculated to be 83.3%.
• The TGNN-based classification demonstrated robust performance in distinguishing between VSD and valvular regurgitation.

Conclusions:

• Intelligent phonocardiography, powered by the TGNN algorithm, shows significant promise for clinical application.
• The system can serve as a valuable decision support tool for pediatric cardiologists.
• Further validation may lead to improved non-invasive diagnosis of congenital heart defects and valvular issues.