Multi-level feature encoding algorithm based on FBPSI for heart sound classification.

Area of Science:

• Cardiology
• Biomedical Signal Processing
• Machine Learning in Healthcare

Background:

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

Purpose of the Study:

• To propose a multi-level feature encoding algorithm using frequency-balanced power spectral intensity for heart sound signal classification.
• To evaluate the algorithm's performance on both public and self-collected cardiac datasets.

Main Methods:

• Denoising heart sound signals using a wavelet threshold function.
• Calculating the frequency-balanced power spectral intensity envelope.
• Extracting multi-level features with an encoder.
• Classifying signals using an ensemble bagging tree classifier.

Main Results:

• Achieved 98.73% average accuracy for normal vs. abnormal heart sound classification.
• Achieved 98.12% average accuracy for normal vs. hypertrophic cardiomyopathy classification.
• Demonstrated robust performance on binary and ternary classification tasks.

Conclusions:

• The proposed multi-level feature encoding algorithm offers a promising approach for accurate heart sound classification.
• This method has significant potential for the early diagnosis of cardiac diseases.
• The findings support the use of advanced signal processing and machine learning in cardiology.