Automated pediatric cardiac auscultation

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence

Background:

• Congenital heart defects require early detection for optimal long-term outcomes.
• Misclassification of heart defects can lead to severe consequences.
• A shortage of physicians in South Africa necessitates accessible diagnostic tools.

Purpose of the Study:

• To develop and evaluate automated techniques for discriminating between pathological and non-pathological heart sounds.
• To assess the efficacy of artificial neural networks, direct ratio, and wavelet analysis for automated cardiac screening.

Main Methods:

• Digitization of auscultation and electrocardiogram (ECG) data from 163 pediatric patients (2 months to 16 years).
• Implementation of an artificial neural network (ANN) model.
• Application of direct ratio and wavelet analysis techniques for sound discrimination.

Main Results:

• The ANN achieved 90% sensitivity and 96.5% specificity using the Jack-knife method.
• Statistical analysis indicates potential for >99% sensitivity and specificity with sufficient training data.
• Comparison of ANN, direct ratio, and wavelet analysis performance in classifying heart sounds.

Conclusions:

• Automated screening using artificial neural networks shows high accuracy in identifying congenital heart defects.
• This technology can significantly benefit rural clinics in South Africa staffed by nurses.
• Further development with more training data can enhance diagnostic performance for pediatric cardiac conditions.