Tunable Q-factor wavelet transform based lung signal decomposition and statistical feature extraction for effective...

Area of Science:

• Medical Informatics
• Signal Processing
• Machine Learning

Background:

• Lung diseases are a leading cause of death globally.
• Traditional auscultation has limitations due to analog stethoscopes and subjective interpretation.
• Automated diagnosis using digitized lung sounds is an active research area.

Purpose of the Study:

• To develop and evaluate a novel computer-based approach for lung disease classification.
• To leverage Tunable Q-factor Wavelet Transform (TQWT) for feature extraction from lung sounds.
• To enhance classification performance using individual and ensemble machine learning models.

Main Methods:

• Feature extraction using statistical measures derived from TQWT decomposition of lung sounds.
• Training individual machine learning models (e.g., SVM, Random Forest) on extracted features.
• Employing ensemble methods (hard and soft voting) to combine predictions from individual models for improved accuracy.
• Evaluating the approach through patient-based and sample-based classification tasks.

Main Results:

• Achieved 97.63% accuracy for binary classification (healthy vs. non-healthy) in patient-based evaluation.
• Obtained up to 66.32% accuracy for three-class and 53.42% for five-class patient-based classification.
• Demonstrated superior performance in sample-based classification compared to existing methods.
• Highlighted the adaptive time-frequency resolution capabilities of TQWT for signal analysis.

Conclusions:

• The proposed TQWT-based feature extraction combined with ensemble learning shows significant potential for automated lung disease diagnosis.
• The approach offers a promising direction for developing more accurate and objective digital auscultation techniques.
• Further research can refine the models for broader clinical application and improved diagnostic accuracy across various lung conditions.