Wheeze and Crackle Discrimination Algorithm in Pneumonia Respiratory Signals

Area of Science:

• Medical Informatics
• Respiratory Medicine
• Machine Learning

Background:

• Pneumonia detection from respiratory sounds is crucial for timely diagnosis and treatment.
• Distinguishing between wheezing and crackling sounds in pneumonia patients aids in targeted therapy.
• Existing methods may lack accuracy or the ability to differentiate specific respiratory sounds.

Purpose of the Study:

• To propose a novel, two-step hierarchical method for pneumonia detection in respiratory sound signals.
• To enhance pneumonia detection performance and add wheeze/crackle discrimination capabilities.
• To facilitate the application of appropriate remedies based on specific respiratory sound characteristics.

Main Methods:

• A two-step hierarchical classification approach was developed.
• Resampling techniques were employed to address data imbalance in the ICBHI pneumonia dataset.
• The random forest algorithm was utilized for both pneumonia classification and wheeze/crackle discrimination.

Main Results:

• The proposed method achieved 85.40% accuracy in detecting pneumonia from respiratory sounds.
• The system demonstrated 82.70% accuracy in discriminating between wheeze and crackle sounds in pneumonia cases.
• The random forest-based hierarchical approach showed improved performance on the ICBHI respiratory dataset.

Conclusions:

• The developed hierarchical random forest method offers an effective approach for pneumonia detection and respiratory sound characterization.
• This method can improve diagnostic accuracy and guide clinical decision-making for pneumonia patients.
• The integration of sound discrimination enhances the clinical applicability of automated respiratory sound analysis.