[Heart sound classification based on improved mel frequency cepstrum coefficient and integrated decision network...

Area of Science:

• Cardiology
• Biomedical Signal Processing
• Machine Learning

Context:

• Early diagnosis of congenital heart disease (CHD) is critical for patient outcomes.
• Traditional heart sound analysis methods often rely on complex cardiac cycle segmentation.
• Accurate and efficient heart sound classification remains a significant challenge in CHD detection.

Purpose:

• To propose a novel, robust heart sound classification method for early CHD diagnosis.
• To enhance feature extraction using an improved Mel Frequency Cepstral Coefficient (MFCC) approach.
• To develop an integrated decision network classifier that avoids cardiac cycle segmentation.

Summary:

• A new method enhances Mel Frequency Cepstral Coefficients (MFCC) with a Fisher discriminant half raised-sine function (F-HRSF) for heart sound feature extraction.
• An integrated decision network, combining Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), classifies heart sounds.
• The system achieves high performance metrics (e.g., 92.15% accuracy) without requiring cardiac cycle segmentation.

Impact:

• The proposed algorithm demonstrates significant potential for the early and accurate diagnosis of congenital heart disease.
• This non-invasive technique could improve CHD screening and diagnostic workflows.
• The enhanced feature extraction and integrated classification network offer a promising advancement in automated cardiac auscultation.