Attention Feature Fusion Network via Knowledge Propagation for Automated Respiratory Sound Classification.

Area of Science:

• Artificial Intelligence
• Medical Diagnostics
• Pulmonology

Background:

• Early diagnosis of respiratory diseases is critical, especially during pandemics.
• Traditional methods like CT and MRI have accessibility limitations.
• Lung auscultation is subjective and requires expert interpretation, further challenged by restricted consultations.

Purpose of the Study:

• To develop an automated respiratory sound classification system using deep learning.
• To enable accurate and remote diagnosis of respiratory conditions.
• To overcome limitations of traditional diagnostic tools and subjective auscultation.

Main Methods:

• Developed a deep convolutional neural network (CNN) model.
• Utilized spectrographic representations of respiratory sounds within an image classification framework.
• Enhanced the CNN with attention feature fusion and a knowledge propagation mechanism.

Main Results:

• The proposed CNN model demonstrated superior performance over state-of-the-art methods.
• The model achieved higher sensitivity in detecting abnormalities in a pediatric dataset.
• Results indicate potential for improved accuracy in respiratory disease diagnosis.

Conclusions:

• Integrating a knowledge propagation mechanism into CNNs advances automated respiratory disease diagnosis.
• This approach offers more accessible and precise healthcare solutions.
• The system is particularly valuable for remote diagnostics during health crises.