A New Method for Heart Disease Detection: Long Short-Term Feature Extraction from Heart Sound Data.

Area of Science:

• Cardiology and Biomedical Signal Processing
• Artificial Intelligence in Healthcare

Background:

• Heart sound analysis is crucial for diagnosing heart disease.
• Traditional methods using fixed-length windows have limitations in capturing complete signal information.

Purpose of the Study:

• To develop an improved feature extraction method for heart sound analysis.
• To enhance the accuracy of machine learning-based heart disease diagnosis.

Main Methods:

• A novel approach integrating short-term (window-based) and long-term (entire signal) features was proposed.
• Features were extracted from five-second fragments and the complete heart sound signal.
• A combined feature pool, termed 'long short-term features', was created for classification.

Main Results:

• The proposed feature extraction method was evaluated using traditional machine learning algorithms on the PhysioNet/CinC Challenge 2016 dataset.
• The integration of multi-scale features demonstrated a significant improvement in diagnostic accuracy.
• Accuracy in heart disease diagnosis was increased by nearly 10%.

Conclusions:

• The novel 'long short-term features' approach enhances the performance of machine learning models for heart sound analysis.
• This method offers a more comprehensive representation of heart sound signals, leading to better diagnostic accuracy.
• The findings suggest a promising direction for improving automated heart disease detection systems.