A novel feature extraction technique for pulmonary sound analysis based on EMD

Area of Science:

• Pulmonary diagnostics
• Biomedical signal processing
• Pattern recognition

Background:

• Stethoscope auscultation is a primary diagnostic tool for chest sounds but is limited by physician experience and signal clarity.
• Automated computer-aided diagnostic systems are needed for reliable pulmonary function analysis, especially in noisy environments.

Purpose of the Study:

• To introduce a novel feature extraction technique for automated discrimination of pulmonary dysfunctions.
• To develop a computer-aided diagnostic system for chest sound analysis using pattern recognition.

Main Methods:

• Identified disease-correlated lung sound signal characteristics using statistical distribution parameters (mean, variance, skewness, kurtosis).
• Extracted features from morphological components of signals in the empirical mode decomposition domain.
• Utilized a classifier model to differentiate between various pulmonary dysfunction classes.

Main Results:

• Validated feature significance through experiments with supervised and unsupervised classifiers.
• Compared the discriminating power of proposed features against three baseline feature types.
• Evaluated experimental results using statistical analysis and physician inference.

Conclusions:

• The proposed feature extraction technique surpasses baseline methods in classification accuracy, sensitivity, and specificity.
• Achieved high performance with an artificial neural network classifier: 94.16% accuracy, 100% sensitivity, and 93.75% specificity.
• The developed method demonstrates superior results across various conditions compared to existing approaches.