PCG Classification Using Multidomain Features and SVM Classifier.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Accurate classification of heart sound recordings is crucial for diagnosing cardiac conditions.
• Existing methods for heart sound analysis often face challenges in feature extraction and classification accuracy.

Purpose of the Study:

• To propose and evaluate a novel method for classifying normal and abnormal heart sound recordings.
• To leverage multidomain features and Support Vector Machine (SVM) for enhanced classification performance.

Main Methods:

• Extraction of 515 features from nine distinct domains including time interval, frequency spectrum, energy, and entropy.
• Correlation analysis to identify the most discriminative features, highlighting frequency spectrum of state, energy, and entropy.
• Training and testing a Support Vector Machine (SVM) classifier with a radial basis kernel function using selected top features.

Main Results:

• The proposed method achieved high average sensitivity (0.88), specificity (0.87), and overall score (0.88) using the top 400 features.
• The SVM classifier demonstrated robust performance even with a reduced number of features for training.
• The classification output remained stable irrespective of the random selection of training features.

Conclusions:

• The combination of multidomain features and SVM classifier is highly effective for heart sound classification.
• The proposed method offers competitive performance compared to previous state-of-the-art scores.
• The approach shows potential for reliable and accurate automated cardiac diagnostics.